Sch2006bp

Sch2006bp
A sublimation technique for
high-precision δ13C on CO2 and CO2 mixing ratio
from air trapped in deep ice cores
Jochen Schmitt
Dissertation zur Erlangung des Grades
Dr. rer. nat.
vorgelegt dem
Fachbereich Geowissenschaften
der Universität Bremen
Bremerhaven, Oktober 2006
Stiftung Alfred-Wegener-Institut für Polar- und Meeresforschung
in der Helmholtz-Gemeinschaft
A sublimation technique for
high-precision δ13C on CO2 and CO2 mixing ratio
from air trapped in deep ice cores
Gutachter:
Prof. Dr. Heinz Miller
Prof. Dr. Kai-Uwe Hinrichs
Stiftung Alfred-Wegener-Institut für Polar- und Meeresforschung
in der Helmholtz-Gemeinschaft
Contents
Zusammenfassung......................................................................................................... v
Thesis summary...........................................................................................................vii
1
Introduction............................................................................................................ 1
1.1
The global carbon cycle ............................................................................................... 3
1.2
Stable carbon isotopes and fractionation processes .................................................. 5
1.3
Reconstructing atmospheric δ13C in the past............................................................. 8
Processes affecting δ13C of the ice core gas archive.......................................... 16
2
2.1
Physical processes....................................................................................................... 16
2.1.1
Firn structure ........................................................................................................ 17
2.1.2
Gas transport properties within the firn column................................................... 18
2.1.3
Enclosure process, age distribution and Δ-age ..................................................... 19
2.1.4
Separation of gases and isotopes due to gravitation............................................. 21
2.1.5
Diffusion along a thermal gradient....................................................................... 22
2.1.6
Diffusion along a concentration gradient ............................................................. 23
2.1.7
Processes during bubble close-off........................................................................ 24
2.1.8
Clathrate formation and disintegration................................................................. 25
2.2
Chemical reactions and a possible in-situ production of CO2 ................................ 28
2.2.1
Indications for CO2 in-situ production from Greenland ice cores........................ 29
2.2.2
In-situ reactions of organic compounds ............................................................... 33
2.2.3
In-situ reactions of inorganic carbonate ............................................................... 38
Conclusions and requirements for the δ13C analysis of ice cores........................... 40
2.3
2.3.1
δ13C artifacts as a matter of scale ......................................................................... 41
2.3.2
Analytical requirements ....................................................................................... 42
ii
Methods and instruments for δ13C and CO2 analysis on ice cores ................. 43
3
3.1
Introduction ................................................................................................................ 43
3.2
Previous approaches for δ13C on ice cores using mechanical devices ................... 43
3.3
Quantitative extraction techniques for ice cores – sublimation in vacuum.......... 46
3.4
General layout of the entire method......................................................................... 49
3.5
The sublimation extraction system ........................................................................... 53
3.5.1
Overall idea behind the sample trapping.............................................................. 53
3.5.2
Vacuum system and water removal ..................................................................... 56
3.5.3
Cooling system for compressed air ...................................................................... 56
3.5.4
Sublimation vessel, internal water trap, and IR lamps......................................... 58
3.5.5
External water trap ............................................................................................... 59
3.5.6
CO2 trap and glass capillaries for CO2 storage .................................................... 62
3.5.7
Molesieve trap - air content measurement ........................................................... 63
3.5.8
Whole air reference inlet to introduce whole air standards.................................. 65
3.6
The tube cracker-GC-IRMS system (CF-IRMS) .................................................... 67
3.6.1
Reasoning for a GC separation of the extracted gas sample ................................ 68
3.6.2
General layout of the measurement sequence for the CF-IRMS system ............. 69
3.6.3
The tube cracker................................................................................................... 72
3.6.4
Device to introduce CO2 reference gas to the tube cracker ................................. 73
3.6.5
Humidifier for the He carrier (GC flow).............................................................. 74
3.6.6
Cryofocus and gas chromatographic separation of CO2 and N2O ....................... 75
3.6.7
Open split and IRMS measurement ..................................................................... 77
3.7
Description on of the analysis procedure ................................................................. 78
3.7.1
Sublimation extraction for ice core samples ........................................................ 78
3.7.2
Procedure to verify the analysis with air standards.............................................. 81
3.7.3
Tube cracker-GC-IRMS measurement scheme ................................................... 82
3.8
Raw data processing and performance of the CF-IRMS analysis......................... 83
3.8.1
δ13C ...................................................................................................................... 84
3.8.2
CO2 concentration ................................................................................................ 87
iii
3.9
Results from air standard and blank measurements .............................................. 89
3.9.1
Reproducibility of δ13C and overall accuracy for air standards ........................... 90
3.9.2
Reproducibility of the CO2 concentration for air standards ................................. 92
3.9.3
Estimation of ‘side effects’ from air standards .................................................... 94
3.9.4
Procedural blanks ................................................................................................. 97
4
Results and discussion of ice core measurements ........................................... 100
4.1
Ice core samples from the EDML ice core ............................................................. 100
4.2
Main characteristics of the measured ice core data .............................................. 102
4.3
Systematic differences during the sublimation extraction ................................... 103
4.4
Small-scale variability – differences among replicates ......................................... 108
4.5
Data comparison of the absolute values ................................................................. 110
4.5.1
Gravitational correction and ice core dating ...................................................... 110
4.5.2
Comparison with other Antarctic ice core records ............................................. 112
4.6
Outlook ...................................................................................................................... 117
References .................................................................................................................. 119
5
Appendix: ........................................................................................................... 129
Acknowledgements.................................................................................................... 165
Erklärung................................................................................................................... 167
iv
Zusammenfassung
Das Eis der polaren Eisschilde stellt das einzige direkte Archiv dar, um Informationen über
die Zusammensetzung der Atmosphäre in der Vergangenheit zu erhalten. Bereits seit mehr als
10 Jahren weiß man durch Analyse von antarktischen Eiskernen, wie z.B. dem Vostok Eiskern, dass die Konzentration des Treibhausgases CO2 zwischen Eiszeiten und Warmzeiten
periodisch schwankte. Während der letzten vier Eiszeiten lag die CO2 Konzentration um etwa
90 ppmv niedriger als zu warmen Klimaperioden. Überraschend war die hohe Korrelation der
CO2 Konzentration mit der Temperatur. Mit Hilfe des Dome C Eiskerns, der im Rahmen des
europäischen Projekts für Eiskernbohrungen in der Antarktis (EPICA) erbohrt wurde, kann
nun die Entwicklung der CO2 Konzentration bis zu 650 000 Jahre in die Vergangenheit zurückverfolgt werden. Die steuernden Prozesse für die markanten CO2 Schwankungen auch
quantitativ zu beschreiben, ist für die Paläoklimaforschung von überragender Bedeutung. Dieses Prozessverständnis ist die Grundvoraussetzung, um Vorhersagen über die zukünftige Entwicklung der atmosphärischen CO2 Konzentration treffen zu können. Eine entscheidende
Schlüsselinformation zur Beantwortung der offenen Fragen, wie atmosphärisches CO2 auf
Veränderungen im globalen Kohlenstoffkreislaufs reagiert, bietet das Verhältnis der stabilen
Kohlenstoffisotope im CO2 (δ13C). Jedoch blieb aufgrund von methodischen Schwierigkeiten
die Analyse von δ13C in Eiskernen auf den oberen Kernabschnitt beschränkt. Darüber hinaus
sind die bisher gewonnen δ13C Daten aufgrund der Messungenauigkeit und ihrer zeitlicher
Auflösung schwierig zu interpretieren. Ferner wurde kürzlich an den EPICA Eiskernen eine
hohe Variabilität des δ13C Signals auf der Zentimeterskala gemessen, die nicht mit atmosphärischen Veränderungen in Einklang zu bringen ist.
Diese Doktorarbeit stellt ein neuartiges Analyseverfahren vor, mit dem es möglich ist, eingeschlossene Luft aus Eiskernen quantitativ zu extrahieren und an dieser δ13C mit Hilfe einer
Kopplung aus Gaschromatographie und Isotopenverhältnismassenspektrometrie in hoher Präzision zu bestimmen. Weiterhin kann mit diesem Verfahren die Konzentration von CO2 an der
gleichen Eisprobe bestimmt werden. Mit dieser neuen Analysetechnik gelingt es, die methodischen Schwierigkeiten der bisherigen Analyseverfahren zu überwinden. Durch Verwendung
einer neuartigen Sublimationstechnik wurde eine nahezu vollständige Extraktionsausbeute
erreicht, was für die genaue Bestimmung von δ13C an tiefen, klathratisierten Eiskernproben
Vorraussetzung ist. Da die Sublimationstechnik für Blaseneis und Klathrateis gleichermaßen
geeignet ist, kann erstmals ein Eiskern in seiner gesamten Länge mit einem einzigen Analyseverfahren auf δ13C und die CO2 Konzentration analysiert werden. Somit erübrigt sich der kritische Schritt, die Extraktionsbedingungen an die Eiszusammensetzung anzupassen. Von einer
Eiskernprobe mit einem Gewicht von ca. 30 g werden während der kontinuierlichen Sublimation fünf einzelne Teilproben separat gesammelt. Dies erlaubt es, fünf getrennte Messungen
pro Eiskernprobe durchzuführen, was einer Probenmenge von ca. 6 g Eis pro Teilprobe ent-
vi
spricht; oder einer extrahierten Luftmenge von etwa 0.5 ml STP. Diese Menge ist um eine bis
zwei Größenordnungen geringer als bei früheren Verfahren zur δ13C Bestimmung. Das
schrittweise Sammeln von fünf Teilproben während der kontinuierlichen Gasextraktion ermöglicht es, den Sublimationsprozess zu überwachen und mögliche unerwünschte Effekte zu
identifizieren. Bisherige Methoden zur δ13C Analyse nutzen mechanische Extraktionsverfahren, bei denen die eingeschlossene Luft in einem einzigen Schritt freigesetzt wird. Folglich
liefern bisherige Verfahren nur Informationen über diesen Analyseschritt und erlauben darüber hinaus keine weitere Prozessinformation. Durch Anwendung einer gaschromatographischen Probenaufreinigung vor der eigentlichen massenspektrometrischen Messung, können
isobare Störkomponenten wie N2O, organische Verbindungen und Verunreinigungen durch
die Bohrflüssigkeit sicher von CO2 abgetrennt werden. Dies war bislang nicht möglich. Die
Richtigkeit und Genauigkeit des Analyseverfahrens wurde mit Hilfe von Luftstandards verifiziert. Diese können der Extraktionsapparatur kontinuierlich zugeführt werden und simulieren
so die Analyse von Eiskernproben so nah wie möglich. Luftstandards können mit einer Genauigkeit von 0.06‰ auf ihren δ13C Wert und mit 1.5 ppmv auf ihre CO2 Konzentration analysiert werden. Die erzielte Leistungsfähigkeit des Verfahrens erlaubt es, die δ13C Änderungen, die mit den beobachteten Änderungen der CO2 Konzentration in der Vergangenheit verknüpft sind, aufzulösen.
An ausgewählten Tiefenintervallen des EPICA Dronning Maud Land Eiskerns wurden Wiederholungsmessungen mit dem Ziel durchgeführt, die kürzlich gefundenen kleinskaligen δ13C
Fluktuationen mit dem neuen Verfahren zu verifizieren. Die hierbei gemessene Reproduzierbarkeit für eng benachbarte Eiskernproben beträgt 0.06‰, folglich die gleiche Streuung, die
auch bei der Analyse von Luftstandards gemessen wurde. Aus diesen Resultaten lässt sich
ableiten, dass die postulierte kleinskalige δ13C Fluktuation im Eis mit diesem Verfahren nicht
bestätigt werden kann. Weiterhin sind die gemessenen CO2 Konzentrationen in guter Übereinstimmung mit publizierten Werten, die mit mechanischen Extraktionsmethoden erzielt
wurden. Die gute Übereinstimmung der CO2 Konzentration mit bisherigen Messungen unterstützt die Eignung der Methode zur richtigen Bestimmung auch von δ13C in Eiskernen. Diese
vorliegende Studie zeigt, dass die neu entwickelte Kopplung aus SublimationsextrationGaschromatographie-Isotopenverhältnismassenspektrometrie geeignet ist, die Änderung in
der δ13C Signatur aufgrund von Verschiebungen im globalen Kohlenstoffkreislauf nachzuweisen.
Thesis summary
Glacier ice from polar ice sheets represents the only direct archive to retrieve information
about the composition of the paleoatmosphere. From deep Antarctic ice cores like Vostok it is
known for more than a decade that the concentration of the greenhouse gas CO2 periodically
varied between glacials and interglacials and its variations are strikingly correlated with temperature. During the glacials the CO2 concentration was about 90 ppmv lower than during the
warm interglacials. With the Dome C ice core drilled within the European Project for Ice Coring in Antarctica (EPICA) the CO2 concentration record now covers the last 650,000 years.
The task to quantitatively understand the processes behind these observed CO2 changes is of
outstanding importance not only for the paleo climate community, but also to predict the CO2
concentration in the future. One crucial key to unravel the open questions about the coupling
of the atmospheric CO2 concentration with the dynamics of the global carbon cycle is the stable carbon isotope ratio of CO2 (δ13C). Methodological constraints have, so far, restricted the
δ13C analysis from ice cores to the uppermost core section. Moreover, the available δ13C data
sets are difficult to interpret as accuracy and temporal resolution are insufficient and a large
centimeter scale variability of δ13C was recently measured on the EPICA ice cores, which
cannot be explained by atmospheric fluctuations.
This thesis introduces a new analytical method allowing a quantitative extraction of air from
small ice core samples coupled to a high precision gas chromatography-isotope ratio mass
spectrometry system to determine δ13C on the released CO2. Additionally, this technique precisely determines the CO2 concentration on the same ice sample. This new technique makes it
possible to surmount major analytical limitations and shortcomings encountered during previous studies. Using a unique sublimation technique a nearly complete gas extraction has been
achieved which represents a prerequisite for high-precision δ13C analysis in deep clathrate ice.
As the sublimation technique is equally suitable to analyze bubble and clathrate ice, this new
method allows for the first time to measure δ13C values and the CO2 concentration on the entire length of an ice core without adapting the extraction to changing ice conditions. From one
ice core sample weighing ~30 g five individual sub-samples are collected separately during
the continuous sublimation to yield five measurements per ice core sample. This corresponds
to ~6 g ice per sub-sample or an amount of roughly 0.5 mL STP of extracted air, i.e. 1-2 orders of magnitude less than in previous conventional ice core δ13C analyses. A stepwise collection of five sub-samples during the ongoing extraction makes it possible to derive valuable
process information during the sublimation. With the previous mechanical extraction techniques applied for δ13C analysis, the enclosed air is extracted all at once, thus, the information
is provided only from one single step. Applying a gas chromatographic sample clean-up prior
to the mass spectrometric measurement, isobaric interferences like N2O, organic compounds,
viii
and drill fluid contaminants can be separated from CO2, which was not possible so far. The
accuracy and precision of the method was verified with whole air reference standards, which
can be introduced to the extraction apparatus and closely mimic the sample analysis. Air standards can be analyzed with a precision of 0.06‰ for δ13C, and 1.5 ppmv for the CO2 concentration. The achieved performance enables to resolve the natural δ13C variability connected
with the observed CO2 concentration changes in the past.
On selected depth intervals on the EPICA Dronning Maud Land ice core sample replicates
were measured to reproduce the recently reported small-scale δ13C scatter. The measurement
reproducibility of δ13C obtained from adjacent ice samples is on average 0.06‰, thus, the
same precision as for air standards. There is no evidence for an assumed small-scale δ13C
variability within the ice itself. Additionally, the CO2 concentrations measured on the same
samples are in line with results from previous studies using mechanical extraction devices.
The achieved accuracy of the CO2 concentration measurement supports the method’s suitability to correctly analyze δ13C in ice cores. As a conclusion of this study, the developed sublimation extraction-gas chromatography-isotope ratio mass spectrometry system is suitable to
trace the paleoatmospheric δ13C changes caused by the dynamics of the global carbon cycle.
1 Introduction
Long before carbon dioxide (CO2) made its way into the current Global Change discussion,
the Swedish chemist Svante Arrhenius already thought about its climatic effect (Arrhenius,
1896). This was decades prior to the rapid atmospheric CO2 rise caused by human activities
and long before measurement techniques were developed to track this rise. He pointed out that
atmospheric gases like CO2 selectively absorb radiation and thus considerably increase the
Earth’s temperature by up to 15 °C. Without this so called natural greenhouse effect the Earth
would not be inhabitable as we know it. At this time, Arrhenius already speculated whether
the ice ages could have been related to reduced CO2 concentrations.
Two iconic figures have shaped the public awareness of the current increase of the atmospheric CO2 concentration. First, the so called ‘Mauna Loa curve’ (Fig. 1-1) from direct atmospheric CO2 measurements on Hawaii, started 1959 under the initiative of Charles D. Keeling documenting its steady rise (Keeling and Whorf, 2000). This curve is unique since it is the
longest data set from direct atmospheric CO2 concentration measurements. It unequivocally
documents the continuous rise from 315 ppmv 1 in 1959 when the measurements were started
to 380 ppmv in 2005.
Atmospheric CO 2 concentration [ppmv]
The second iconic figure is the ‘Vostok record’ (Petit et al., 1999) allowing to look back at
380
370
360
350
340
330
320
1960
1970
1980
Year [AD]
1990
2000
Figure 1-1 Increase of atmospheric CO2 since first measurements started at Mauna Loa station, Hawaii
(Keeling and Whorf, 2000). During the last 45 years the CO2 concentration rose from 315 to 380 ppmv with an
average annual increase of 1.4 ppmv year-1. This translates to roughly 7 Pg 2 CO2 per year added to the atmosphere by direct (fossil fuel burning) or indirect (land use change) emissions. Besides this long term trend attributed to human activity, the record shows an annual cycle, which reflects the seasonal imbalance of the net carbon uptake and release by the terrestrial biosphere.
1
2
1 ppmv = 1 part per million by volume or mole fraction
Pg =1015 grams and equivalent to Gigatons (Gt)
2
Introduction
four glacial cycles of CO2 and climate history (Fig. 1-2). Within the perspective of the last
400,000 years it became obvious that the current level of 380 ppmv was unprecedented. The
mean value of the last 10,000 years (corresponding to the Holocene) was 270 ppmv. During
ice ages the CO2 level was considerably lower and dropped to almost 180 ppmv during the
coldest periods. The ‘Vostok record’ shows an intimate coupling of the CO2 concentration
and δD 3, which is a proxy 4 for the Antarctic temperature. These two pieces of information
certainly influenced society and politics and boosted international efforts like the IPCC (Intergovernmental Panel on Climate Change) and the Kyoto protocol trying to slow down the
CO2 increase in the future.
CO2 [ppmv ]
For the scientific community the task remained to explain the enigmatic glacial/interglacial
300
250
-400
200
-450
-475
δ D [‰]
-425
Sea lev el [m]
0
-50
-100
0
50
100
150
200
250
300
350
400
Age [ka BP]
Figure 1-2 Compilation of the Vostok CO2 concentrations and δD as a proxy for local air temperature (Petit et
al., 1999) and the changes in global sea level relative to the present level (Bintanja et al., 2005). To a first approximation, sea level changes reflect the volume of ocean water bound in continental ice sheets during the ice
ages. Note that in this figure time is on a the geologic time scale and therefore runs from right to left, thus,
opposite to the direction in Figure 1-1; time is given in years before present (BP), which is defined at 1950
Anno Domini (AD). CO2 minima were reached approximately when the sea level was at a minimum, hence, the
extent of the continental ice sheets were at a maximum. Vice versa, highest CO2 levels were found during interglacials during the high stands of the sea level. The evolution of the local temperature (as deduced from δD)
follows this overall picture and points to a strong coupling of the climate and the carbon cycle. On a finer time
scale leads and lags and a non-linear behavior of the system is clearly visible.
3
δD is the isotopic ratio of 2H (or Deuterium) to 1H measured on H2O to derive a measure of the precipitation
temperature at a site (δD is often used as equivalent to δ18O).
4
A proxy is a measurable parameter, i.e. δD in precipitation, from which a target environmental information like
annual temperature can be deduced.
Introduction
3
CO2 riddle and to extend the records further back in time. The new results have added well
resolved CO2 data of the next two older glacial/interglacial climate cycles to the four cycles
already investigated from the Vostok ice core. These two cycles are characterised by considerably lower temperature maxima during interglacials, but with similar temperatures during
glacial periods. As the same climate-CO2 relation was found valid even under these quite different climate cycles, the established picture of a close climate-CO2 relation was strongly confirmed.
Within the scientific community a broad consensus exists about the temporal evolution of the
past’s atmospheric CO2 concentration. In contrast, for the underlying processes and feedback
loops and the question of what drives what, things are different (Broecker and Peng, 1987;
Toggweiler, 1999; Archer et al., 2000; Köhler et al., 2005a). Only recently, Köhler et al
(2005a) were able to propose a global carbon cycle scenario, which is able to quantitatively
account for the observed CO2 changes. While the different processes contributing to the glacial/interglacial CO2 change seem now to be recognized, their individual contribution, however, may vary depending on the scenario.
Another, additional constraint on the carbon cycle dynamics is provided by studies on the
isotopic signature of CO2 in the past, which so far remained rudimentary (Friedli et al., 1986;
Indermühle et al., 1999; Smith et al., 1999; Fischer et al., 2003). This thesis presents a new
methodological attempt to gain new high precision δ13C data on CO2 from deep ice cores.
Understanding atmospheric CO2 concentration on glacial/interglacial timescales is challenging, and also a prerequisite to assess its future behaviour, which is on top of the Global
Change agenda.
1.1 The global carbon cycle
The main characteristic of atmospheric CO2 is its rapid exchange with both the ocean and the
terrestrial biosphere. Note that the carbonate sediments of the Earth crust constitute the largest
carbon reservoir. Although its reservoir size of 48,000,000 Pg C dwarfs the ocean (39,000 Pg
C), the exchange rate due to weathering on land or sedimentation/dissolution balance within
the ocean are relatively small (both around 0.2 Pg C year-1) compared to the ocean and biosphere fluxes. Nevertheless, the interplay of carbonate precipitation in the surface ocean and
its dissolution in the abyss constitutes the carbon and alkalinity equilibrium of the ocean on
longer time scales. To bring the three most dynamic carbon reservoirs (ocean, biosphere, and
atmosphere) into relation, a first step is to compare the corresponding reservoir sizes. The
reservoir of the terrestrial biosphere, where carbon is stored mainly in soils and plant biomass,
is by a factor of 3 larger than the carbon content of the preindustrial atmosphere. In case of the
ocean, with its main inorganic species bicarbonate and carbonate, the carbon reservoir size is
4
Introduction
by a factor of 65 larger than the atmosphere (Sigman and Boyle, 2000). Physical, chemical
and biological processes link the biosphere and the ocean with the atmosphere and determine
its budget. The main processes and fluxes between these reservoirs are illustrated in a simple
sketch of the global carbon cycle (Fig. 1-3).
The largest annual carbon gross flux occurs between the terrestrial biosphere and the atmosphere, which amounts to around 120 Pg. Prior to the recent human influence, the two opposing fluxes, photosynthesis and respiration/degradation, were almost balanced. The second
largest flux is the gas exchange of the ocean’s surface with the atmosphere, where the outgassing of CO2 from the warm tropical waters is in balance with the net uptake into cold waatmosphere
70 Pg a-1
120 Pg a-1
(- 7‰ )
590 Pg
surface DIC
layer ( 2 ‰ )
marine
50 Pg a-1
38,000 Pg
biota
(-20 to -35‰ )
3 Pg
biological pump
lysocline
CaCO3 + Corg
0.2 Pg a-1 sedimentation and dissolution
terrestrial vegetation
C3
C4
(-28‰ )
(-12 ‰ )
2300 Pg
soil
CaCO3 weathering
0.2 Pg a-1
CaCO3 in marine and continental sediments
( 0‰ ) 48,000,0000 Pg
Figure 1-3 A simple sketch of the global carbon cycle showing the earth’s crust main reservoirs, their annual
exchange fluxes, and in brackets their isotopic composition (δ13C). Although sedimentary calcium carbonate
(CaCO3) constitutes the largest carbon pool, the exchange rate with the three more dynamic reservoirs (ocean,
biosphere and atmosphere) is marginal. In contrast, the smallest carbon reservoir, the atmosphere rapidly exchanges with both the surface ocean and the terrestrial biosphere. The carbon flux associated with the terrestrial
biosphere is highly variable on a seasonal time scale with a net uptake during the main growing season during
the northern summer and CO2 release between fall and spring (see Fig. 1-4 for details). In contrast, the exchange fluxes ocean/atmosphere are regionally separated with the tropics releasing CO2 while the high northern
latitudes are a CO2 sink. Within the ocean the marine biosphere drives the biological carbon pump transferring
organic and inorganic carbon to deeper ocean layers. The oceanic circulation (grey arrow) again redistributes
the accumulated carbon and brings it back to the surface ocean/atmosphere system. Basically, the ratio of the
carbon export from surface to the deep ocean by biological and physical processes to the strength of carbon
redistribution by circulation sets the atmospheric CO2 and δ13C level. Each of the main reservoirs is characterized by a distinct carbon isotopic composition, which results from isotopic fractionation due to fluxes among
the reservoirs. Arrows denote annual fluxes.
Introduction
5
ters at high latitudes. Within the surface ocean the marine biota produces a mixture of organic
carbon rich soft tissue biomass and inorganic carbonate shells. While the estimated actual
carbon stock is only 3 Pg, this relatively small amount of marine biota has a high turnover rate
and the marine annual production of organic carbon sums up to ~50 Pg. A large proportion is
rapidly mineralized within the surface water and only ~10 Pg are exported to deeper layers
(Schlitzer, 2002). On its way down from the productive surface layer to the deep sea, mineralization of organic carbon and dissolution of inorganic carbonate takes place. The biologically induced carbon export from the surface ocean to deeper layers is called the biological
pump. Its action leads to an effective enrichment of dissolved inorganic carbon (DIC is the
sum of dissolved CO2, HCO3 -, and CO32-) with increasing water depth.
A marked geochemical feature of the deep ocean is the lysocline, the water depth where calcium carbonate becomes undersaturated and the progressive dissolution of the calcareous
shells sets in. The action of the global ocean currents (the thermohaline circulation) balances
the successive accumulation of carbon in the deep ocean as the enriched deep water is transported to upwelling regions, which release CO2 back to the atmosphere. Modelling experiments by Brovkin et al. (2002) showed that the marine biologic cycle in total has a capacity to
mediate about 2400 Pg carbon by remineralisation of organic carbon.
1.2 Stable carbon isotopes and fractionation processes
As shown in Figure 1-3 the reservoirs not only exchange carbon in large quantities, but each
reservoir has a distinct carbon isotopic composition. This is due to the fact that exchange
processes between one reservoir and another discriminate between the carbon isotopes. Carbon has two stable isotopes:
12
C which is the most abundant with 98.9% and
13
C with an
abundance of only 1.1%. The isotopic variations measured in natural samples are rather small,
therefore differences in the isotopic ratios are usually reported in permil (‰) relative to a
standard using the δ-notation:
⎡ Rsample − Rstandard
Rstandard
⎢⎣
δ 13C = ⎢
⎤
⎥ ⋅ 1000 [‰]
⎥⎦
(1 − 1)
with
R=
13
C
12
C
and
Rsample : isotopic ratio of the sample
Rstandard : isotopic ratio of the standard
(1 − 2)
Introduction
6
The current reference standard for stable carbon isotope analysis is the Vienna Pee Dee Belemnite (VPDB). As the original reference material from the Belemnite of the Pee Dee Formation in South Carolina (PDB) was exhausted, its 13C/12C ratio of 0.011237 measured by Craig
(1957) was assigned to the new VPDB standard NBS19 (Allison et al., 1995).
Fluxes from one reservoir to another, e.g. from A to B, are always connected with an isotopic
fractionation or discrimination. Fractionation among isotopes are either induced kinetically or
during equilibrium processes (Young et al., 2002). Kinetic fractionations are often related to
physical transport effects and are generally observed during diffusion, whereby the lighter
molecules diffuse faster than the heavier ones. Equilibrium fractionations are involved among
chemical species, e.g. between CO32-, HCO3- and dissolved CO2, and during phase changes,
e.g. gas-liquid equilibrium. The fractionation factor α is defined as
α
A→ B
=
RA
RB
(1 − 3)
As natural fractionation processes are generally small, the absolute number of α is near 1,
thus, a more convenient way to report isotopic fractionation uses ε instead of α.
ε = (α − 1) × 1000 [‰]
(1 − 4)
In both cases the degree with which two or more isotopes are fractionated is mass dependent.
In rare cases, like photochemical reactions of ozone in the stratosphere, mass independent
fractionations occur and influence the oxygen isotope distribution of O3, CO2 and N2O (e.g.
Röckmann et al., 2001). During most natural processes kinetic and equilibrium fractionations
occur simultaneously. As isotope effects are additive, it is convenient to calculate the net fractionation at certain conditions. For many processes the dependencies of the isotopic fractionation on temperature, pressure and other parameters are either calculated by thermodynamics
or by empiric relations.
In case of the primary production of terrestrial plants, the additive effect of many consecutive
steps during the photosynthesis discriminates the heavier isotope 13C against 12C. As a consequence, plants are isotopically depleted with respect to the heavier isotope. The degree with
which plants discriminate carbon isotopes is primarily species dependent (C3, C4 and CAM 5
pathway (O'Leary, 1981)), but also influenced by environmental parameters. The overall
process of photosynthesis of C3 plants discriminates against 13C with a fractionation factor of
about -20‰ (Lloyd and Farquhar, 1994). Given an atmospheric δ13C value of -8‰, the biomass of C3 plants as a typical δ13C value of -28‰. As the removed carbon has a light signa5
The abbreviations C3 and C4 denote the number of carbon atoms of the first intermediate product built during
the photosynthesis. CAM is the abbreviation of Crassulacean Acid Metabolism.
Introduction
7
ture, in turn, the remaining CO2 in the atmosphere gets isotopically enriched in 13C. The basis
for this is the isotopic mass balance, which can be formulated in its simplest form as a two
reservoir system as follows:
C mixture ⋅ δ 13 C mixture = C reservoir A ⋅ δ 13 C reservior A + C reservoir B ⋅ δ 13 C reservior B
(1 − 5)
with Ci the amount of carbon for the reservoirs A or B and the mixture, and δ 13Ci the respective isotopic composition of the reservoirs and the composition of the mixture.
The effect of isotopic fractionation during photosynthesis and the mass balance principle can
be observed from the plant level to the landscape level, and most strikingly, on the global
scale as shown in Figure 1-4. Basically, the annual waxing and waning of the terrestrial biosphere can be followed at three measurement stations. As most of the primary production
from plants is located in the high latitudes of the Northern Hemisphere, the annual carbon
cycle is most pronounced at Barrow station (71°N) compared to Mauna Loa (19°N) and South
Pole (90°S). During peak photosynthesis in summer, large amounts of CO2 are removed from
the atmosphere visible by the marked dip in the CO2 concentration (lower panel). In parallel,
the atmosphere gets isotopically enriched by almost 0.8‰ (upper panel) as predicted by the
mass balance. The annual gross fluxes of the terrestrial photosynthesis amount to ~120 Pg
carbon (Ciais et al., 1997). Due to this rapid turnover of the atmospheric CO2 a δ13C signal
can be quickly diluted within the larger carbon pools, especially in the surface ocean.
Contrary to the isotopic fractionation between the terrestrial biosphere and the atmosphere, for
the ocean/atmosphere exchange it is the ocean, which is ‘heavier’ in δ13C and the atmosphere,
which is ‘lighter’ (see Fig. 1-3). Here, the temperature dependent fractionation of the carbon
isotopes is dominated by the carbonate chemistry of the ocean and the diffusive transfer of
CO2 from air to water and vice versa (Mook, 1986). In contrast to the pronounced seasonal
exchange fluxes of the terrestrial biosphere/atmosphere system, the seasonality of the
ocean/atmosphere exchange fluxes is less pronounced. However, on time scales of decades
and millennia, it is the ocean who dictates the atmospheric CO2 level and its δ13C value. For
the latter, the main factors are the temperature dependent isotopic fractionation during the
ocean/atmospheric gas exchange. Secondly, the vertical δ13C gradient within the ocean, which
is introduced by the biological pump and depends on the quantity of the marine export production and the δ13C value of the exported biomass.
From this simplistic view on the global carbon cycle it can be deduced that knowing the temporal changes of a reservoir’s size and its isotopic composition can provide valuable information about the underlying processes and fluxes. In other words, if we have access to archives
Introduction
8
storing these information, we can identify and quantitatively describe the mechanisms driving
the major CO2 fluctuations in the past (e.g. during glacial/interglacial cycles).
-7.6
13
δ C [‰]
-7.8
-8.0
-8.2
CO 2 concentration [ppmv]
-8.4
380
2001
2002
2003
370
360
2000
Barrow
(71.32' N, 156.60' W )
Mauna Loa (19.32' N, 155.34' W )
South Pole (89.98' S, 24.48' W )
2001
2002
Year [AD]
2003
2004
Figure 1-4 Time series of direct atmospheric δ13C and CO2 concentration measurements at three stations covering four years to illustrate the isotopic fractionation during photosynthesis and the concept of the isotopic mass
balance. The northernmost station, Barrow, is characterized by the highest annual amplitudes: 0.8‰ for δ13C
and 15 ppmv for the atmospheric CO2 concentration. This pattern reflects the net carbon fixation by plants,
which are mainly located in the Northern Hemisphere and the imprint of the associated isotopic fractionation.
As the maxima of the net carbon fluxes of photosynthesis and respiration are not synchronous, the isotopic
fractionation during carbon fixation perturbs the atmosphere δ13C value. Going south, i.e. away from the main
source of the signal, the seasonal amplitude at the Mauna Loa station is already attenuated. At the South Pole,
the seasonal amplitude has almost vanished since limited interhemispheric mixing prevents the northern signal
to penetrate further southwards. The Southern Hemisphere itself lacks large land masses with equivalents to the
boreal forest biomes like in Eurasia. Nevertheless, a phase lag of 6 months visible for the δ13C maxima reflects
the imprint of the southern biosphere on the atmosphere of the Southern Hemisphere. Large CO2 gross fluxes
between the atmosphere and the surface ocean further reduce the amplitude. The general trend towards more
negative δ13C values and increasing CO2 concentration reflects the anthropogenic CO2 emissions of isotopically
depleted carbon. The data shown are taken from the NOAA/CMDL 6 monitoring station network (Pieter Tans,
NOAA, http://www.cmdl.noaa.gov).
1.3 Reconstructing atmospheric δ13C in the past
Since decades carbon isotopic information has been available from the ocean as the largest of
the three dynamic carbon reservoirs (e.g. Curry et al., 1988; Spero and Lea, 2002; Broecker
and Clark, 2003; Hodell et al., 2003). These information were derived from carbonate shells
6
Climate Monitoring and Diagnostics Laboratory of the National Oceanic and Atmospheric Administration
Introduction
9
from a large suite of marine sediment cores, which further allow to reconstruct a broad spectrum of physical, chemical, and biological parameters of the ocean in the past (δ18O, alkenones, boron isotopes, salinity etc.). Therefore, physical oceanographers and the associated
modelling community proposed several hypotheses how the ocean might have controlled the
atmospheric CO2 concentration in the past (e.g. Broecker and Peng, 1987; Toggweiler, 1999;
Archer et al., 2000; Sigman and Boyle, 2000; Stephens and Keeling, 2000; Köhler et al.,
2005a).
Only recently, Ruddiman (2003) brought up the hypothesis that mankind might have influenced the atmospheric CO2 concentration already for thousand of years due to substantial land
use changes. The point of this discussion is that the level of the CO2 concentration throughout
the pre-industrial Holocene was not constant, but dropped from an early maximum of almost
270 ppmv around 11 ka BP to a minimum of 260 ppmv at 8 ka BP (Indermühle et al., 1999;
Monnin et al., 2001; Monnin et al., 2004). Afterwards, CO2 was continuously rising with a
minor decrease during the period of the ‘Little Ice Age’ (around 15th to 18th century) and
eventually sharply increased after the industrial revolution at around 1750 AD (Siegenthaler
et al., 2005a). Interestingly, this unexplained rise in the preindustrial CO2 concentration is
unique for the Holocene. For older interglacials (marine isotope stages (MIS) 5, 7, 9) this rise
was not observed in the CO2 ice core records (Petit et al., 1999; Siegenthaler et al., 2005b).
An intense discussion followed whether this CO2 increase was already induced by human
activities, by natural changes in the terrestrial biosphere, or whether an internal equilibration
of the ocean/sediment system was responsible (e.g. Joos et al., 2004; Claussen et al., 2005;
Ruddiman, 2005; Broecker and Stocker, 2006). Broecker and Clark (2003) summarized arguments in favor of the ocean as the dominant driving mechanism, but left the answer open by
stating: “Only when a convincing 13C record for atmospheric CO2 has been generated will it
be possible to make this distinction.”
Attempts to obtain such a record either from proxies or direct ice core archives have been
made for more then 30 years. Beginning in the 1970s, δ13C was measured on tree rings from
C3 plants to trace back the atmospheric δ13C depletion caused by fossil fuel burning and anthropogenic land use changes (Francey, 1981; Arens et al., 2000). Due to the large isotopic
fractionation of plants using the C3 pathway, the data was highly scattered and not reliable
due to vital and climatic effects influencing the photosynthetic discrimination (Francey and
Farquhar, 1982). C4 plants are principally more suitable to obtain an unbiased atmospheric
δ13C record. This is due to the fact that the C4 pathway discriminates carbon isotopes only
little. Based on this idea, Marino et al. (1992) presented δ13C values from plant remains of a
single C4 species. However, this assumption was challenged (Buchmann et al., 1996 and references therein) and the attempt to reconstruct atmospheric δ13C from plant proxies was abandoned.
10
Introduction
An inherent drawback for using terrestrial δ13C proxies becomes apparent when looking at the
large seasonal δ13C cycles of the atmosphere itself (shown in Fig 1-4). The seasonal δ13C amplitude at potential locations for a terrestrial δ13C proxy is of the same order of magnitude
(~0.5‰) as the entire δ13C variability expected for glacial/interglacial time scales (Smith et
al., 1999, see Fig. 1-7). As the net carbon fixation of plants is highly variable throughout the
year, any climatic shift influencing this temporal distribution of carbon fixation will induce a
bias in the atmospheric δ13C data. The prerequisite of a reliable atmospheric archive spanning
large time scales is that the stored information represents an unbiased value of the past’s atmospheric composition. This implies that the underlying process with which an ideal archive
records annual mean values is not a function of short term fluctuations. Such conditions are
realized for the air bubble archives in ice cores that constitute the only direct archives for atmospheric gases. CO2 trapped in ice cores can be considered representative of the whole troposphere of at least the hemisphere in which the drill site is located. As shown above (Fig. 14) the seasonal δ13C fluctuations for high southern latitudes (station South Pole) are largely
attenuated. Further, air mixing by diffusion within the firn column and the slow bubble enclosure process removes any short term fluctuations and the enclosed air in the ice core represents air samples with an age distribution of many decades.
During the last 25 years analytic efforts were not only made to analyze the CO2 concentration
of the trapped air in ice cores, but also on its carbon isotopic composition (Friedli et al., 1984;
Friedli et al., 1986; Leuenberger et al., 1992; Francey et al., 1999; Indermühle et al., 1999;
Smith et al., 1999; Leuenberger et al., 2003; Eyer, 2004). However, problems within the analyzed ice itself and analytical hurdles have prevented scientists so far to generate long and
well resolved high precision δ13C data sets like the ones available for the CO2 concentration.
The early δ13C measurements on the Siple Dome ice core from Friedli et al. (1986) cover
only the last two centuries (Fig. 1-5). However, this record provided strong evidence that the
recent rise of the atmospheric CO2 concentration is related to carbon emissions carrying an
isotopically ‘light’ δ13C signal (~-25‰). This δ13C signature matches both the carbon isotopic
composition of fossil fuels and the organic carbon found in terrestrial soils and plants. Both
sources emit carbon in large quantities by human activities either directly due to burning of
fossil fuels or indirectly due to land use changes (e.g. conversion of forests into crop lands).
Later measurements on the Law Dome ice core were able to establish a well defined preindustrial atmospheric δ13C level of around -6.4‰ (Francey et al., 1999). The large amount of isotopically light anthropogenic carbon emitted during the last 200 years is diluting the original
atmospheric δ13C signal. The simultaneous decrease in δ13C with increasing CO2 concentrations caused by the addition of a light carbon source to the atmospheric carbon pool is called
Introduction
11
the Suess effect 7. Prior to the anthropogenic CO2 increase starting around 1750 AD, the
global carbon cycle was relatively stable according to the Law Dome record (Fig. 1-5) showing only minor variability in δ13C and in the CO2 concentration (see lower panel of Fig. 1-6
for a wider perspective of the Holocene CO2 evolution).
The first δ13C data from the glacial period were retrieved from the Byrd ice core (Leuenberger
et al., 1992), but the measured data set does not provide the temporal resolution to allow a
reconstruction of the carbon cycle of the glacial period (Fig. 1-7). Nevertheless, the Byrd data
showed that the δ13C signature of the glacial atmosphere was on average 0.3±0.2‰ isotopically lighter than during the Holocene (Leuenberger et al., 1992). This is in line with the general assumption that during the cold periods the carbon reservoir of the terrestrial biosphere
was reduced as large areas in the Northern Hemisphere were covered with continental ice
sheets or permafrost. The amount of organic carbon stored in these areas during the interglacials is assumed to be released into the atmosphere/ocean system, thus, decreasing its δ13C
Age [AD]
13
δ C [‰]
1750
-6.2
-6.4
-6.6
-6.8
-7
-7.2
-7.4
1350
1150
950
800
1000
Law Dome
Siple Dome
330
CO 2 [ppmv]
1550
320
310
300
290
280
200
400
600
Age [years BP]
Figure 1-5 Compilation of δ13C values and CO2 concentrations from two Antarctic ice cores covering the last
1000 years (top panel showing δ13C and the lower panel the respective CO2 concentration measured on the
same core). Note the two different time axes (AD and BP) valid for both records. The results from the Siple
Dome core (Friedli et al., 1986) were measured on large ice samples of ~700 g with a precision of 0.1‰ for
δ13C and 3 ppmv for the CO2 concentration and the records starts with the onset of the marked CO2 rise. More
precise and covering also the preindustrial period is the Law Dome record (Francey et al., 1999) with a precision of mostly 0.025-0.05‰ for δ13C and 1.2 ppmv for the CO2 concentration for ~1kg ice samples.
7
named after H.E. Suess. The Suess effect denotes the dilution of the atmospheric 14C and 13C concentration due
to the emission of ‘old’ carbon from fossil fuels, devoid of 14C and as also isotopically depleted in δ13C.
Introduction
12
signature. A more detailed picture emerged from the two δ13C data sets measured on the Taylor Dome ice core, showing millennial scale δ13C variability within the Holocene (Indermühle
et al., 1999) and especially during the deglaciation between 18 and 10 ka BP (Smith et al.,
1999) summarized in Figure 1-6 and 1-7. Due to the low temporal resolution and limited analytical precision of 0.085‰ many questions remained unanswered. To improve the temporal
resolution within the Holocene and extending the δ13C record further into the glacial period,
recently Eyer (2004) measured two highly resolved δ13C records on the EPICA ice cores
(Dronning Maud Land: EDML; Dome C: EDC). As shown in Figure 1-6, the highly resolved
EDML Holocene δ13C record (Eyer, 2004) deviates from the δ13C values from Taylor Dome
(Indermühle et al., 1999). Even when large dating errors between the two ice cores were taken
into account, it is difficult to align these records. Especially for the time interval older than 6
ka BP the δ13C average for EDML is ~0.3‰ isotopically lighter compared to the Taylor
Dome data, however the analytical scatter in the Eyer (2004) data is quite considerable.
Slightly better is the agreement of the Holocene part of the EDC δ13C record with Taylor
Dome shown in Figure 1-7.
Based on this overview on the previous δ13C data sets, two major issues need special attention: First, the δ13C ice core records cover only the Holocene and the younger part of the gla-
CO 2 [ppmv]
13
δ C [‰]
-6
-6.2
-6.4
-6.6
-6.8
-7
-7.2
-7.4
Law Dome
EDML
Taylor Dome
320
Law Dome
EDC
Taylor Dome
300
280
260
2
4
6
Age [ka BP]
8
10
Figure 1-6 Compilation of δ13C values and CO2 concentrations covering the Holocene period. Top panel: δ13C
data from Law Dome, (black circles, Francey et al., 1999), EDML (blue diamonds, Eyer, 2004), Taylor Dome
(green circles, Indermühle et al., 1999). Lower panel: CO2 records from Law Dome (black circles, Francey et
al., 1999), EDC (grey diamonds, 2004), Taylor Dome (green circles, Indermühle et al., 1999).
Introduction
13
cial period. Secondly, discrepancies of the absolute δ13C values are visible among the different data sets and especially the EDML record shows a high variability on very short time
scales.
The first issue results from the fact that all previous studies used mechanical devices to extract
the air trapped in the ice core samples. Usually, only the upper 700 meters of an ice core,
where air exists in bubbles, are unproblematic for mechanical extraction devices. At deeper
layers, air is tightly bound within the ice crystal and forms clathrate hydrates from which the
gases are difficult to extract. Although Eyer (2004) conducted δ13C measurements on this so
called clathrate ice on the EDC core, the obtained results were more scattered (see EDC δ13C
-6
-6.2
13
δ C [‰]
-6.4
-6.6
-6.8
-7
Law Dome
EDC
Taylor Dome
Byrd
-7.2
-7.4
-7.6
320
Law Dome
EDC
Taylor Dome (1999)
Taylor Dome (2000)
CO 2 [ppmv]
300
280
260
240
220
200
180
2
6
10
14
18
22
Age [ka BP]
26
30
34
38
Figure 1-7 Compilation of δ13C data and CO2 concentrations reaching back to the glacial period. Top panel:
δ13C data from Law Dome (black circles, Francey et al., 1999), EDC (blue diamonds, Eyer, 2004), Taylor
Dome (green circles, Indermühle et al., 1999; Smith et al., 1999) and from the Byrd ice core (yellow circles,
Leuenberger et al., 1992). Lower panel CO2 records from Law Dome (black circles, Francey et al., 1999),
EDC (grey diamonds, Monnin et al., 2001; Monnin et al., 2004), ‘Taylor Dome (1999)’ (green circles, Indermühle et al., 1999; Smith et al., 1999) is the corresponding CO2 data to the δ13C values, and the open green
circles the ‘Taylor Dome (2000)’ data measured by Indermühle et al. (2000) extending the record further back
in time.
Introduction
14
data in Fig. 1-7 for the time period older than 12 ka BP). He reported isotopic effects during
the extraction, which were ascribed to the nonquantitative gas recovery. Since the fractionation process associated with incomplete recovery is not yet understood, the scatter increased
with depth (Fig. 1-7). As almost 90% of the entire climatic history of an ice core is enclosed
in clathrate ice, a quantitative extraction technique for the δ13C analysis is therefore compulsory.
Secondly, while the δ13C data recently measured on the EDC core (Eyer, 2004) generally
agree with the Taylor Dome record (Indermühle et al., 1999; Smith et al., 1999) as shown in
Figure 1-7, the EDML δ13C record (Eyer, 2004) has little resemblance with the Taylor Dome
record (Fig. 1-5). Remarkable is the higher scatter of the EDML δ13C record with 0.23‰
compared to 0.12‰ for the EDC record (Eyer, 2004). This is puzzling since the latter results
were obtained from the same laboratory using the identical measurement technique and protocol. Hence, problems due to differences in the applied analysis technique, the referencing and
correcting scheme should be excluded. Consequently, either the observed δ13C difference between EDML and EDC lies in the ice itself, which would somehow challenge the approach to
yield accurate δ13C values from ice cores, or an up to now undetected analytical effect might
account for this discrepancy. Without exception, all previous studies reported on the issue that
organic impurities (e.g. from drill fluid) disturb the mass spectrometric measurement during
sample analysis. In addition, different approaches were chosen to correct for the isobaric interference of N2O in the mass spectrometric analysis, which affects δ13C by a few tens of ‰.
Based on the experiences from the previous approaches to measure δ13C on ice core samples,
the objective of this work was to develop a new analysis technique to fulfil the following criteria:
•
Quantitative extraction of air from the entire length of an ice core
•
Use of a gas chromatographic sample clean-up to exclude interferences due to contamination from drill fluid
•
Separation of N2O from CO2 prior to the mass spectrometric measurement
•
Drastic reduction of the sample size
•
Parallel determination of the CO2 concentration at a high precision (<2 ppmv) on the
same ice sample to identify a possible contamination within the ice itself or during the
analysis
•
A precision of ~0.05‰ to resolve the small variations connected with the global carbon cycle dynamics
Introduction
15
This work represents the first effort to fulfil all these criteria. To this end, a new sublimation
extraction for δ13C on CO2 in ice core samples coupled to a gas chromatography-isotope ratio
mass spectrometry method has been developed. This new methodology, its analytical set-up
and uncertainty as well as first measurements on Antarctic ice core samples are described
within this work.
This work is structured as follows: In Chapter 2 physical and chemical processes are described which lead to a deviation of the measured δ13C composition compared to the atmospheric value. Chapter 3 presents the development and performance of the analysis system for
measuring δ13C and the CO2 concentration on ice core samples. In Chapter 4 first ice core
results of this new analysis technique are shown and compared with previous ice core studies.
In the Appendix a paper entitled ‘On the application and interpretation of Keeling plots in
paleo climate research – Deciphering δ13C of atmospheric CO2 measured in ice cores’ is presented. This paper is a collaborative work with Peter Köhler and Hubertus Fischer and is
submitted to Biogeosciences.
2 Processes affecting δ13C of the ice core gas archive
In the following chapter known processes are summarized which may lead to a deviation of
the measured δ13C compared to the atmospheric values during the time the air was trapped
within the ice. In order to ultimately interpret the measured δ13C data in terms of changes of
the global carbon cycle, these effects have to be identified and quantified.
These processes mainly fall into two classes: First, physical processes within the firn column
and during the close-off process, as well as the formation of clathrate-hydrates from air bubbles. Secondly, chemical reactions adding extra CO2 after the air was trapped within the ice.
In contrast to the physical processes, this so called in-situ production of CO2 is less understood and as yet cannot be easily corrected for as is the case for the physical ones (Anklin et
al., 1995; Barnola, 1999; Tschumi and Stauffer, 2000).
2.1 Physical processes
Most physical processes relevant for CO2 and δ13C measured on ice cores are related to molecular transport processes occurring in the firn, i.e. gravitative settling, thermal diffusion, and
ordinary diffusion. In the first paragraphs, the basic structural properties of this special medium are introduced since they constitute the boundary conditions for these diffusion processes. On this basis, each diffusive process is then treated separately. However, these processes are interrelated in that both gravitative settling and thermal diffusion create concentration gradients, whereas ordinary diffusion attenuates these gradients until a steady state equilibrium is established. Formally, this is expressed in the following differential equation combining these driving forces (Severinghaus et al., 2001):
∂C ∂ ⎛
⎡ ∂C
= ⎜⎜ Deff (z , T ) ⎢
∂t ∂z ⎝
⎣ ∂z
−
Δmg
RT
+ Ω
dT ⎤ ⎞
⎟
dz ⎥⎦ ⎟⎠
(2 − 1)
1
424
3 14243 14243
ordinary
diffusion
gravitative
settling
thermal
diffusion
with C the isotopic composition of a gas species (e.g. δ13C), t the time, z the depth, Deff the
effective molecular diffusivity of a gas species in firn, and T the temperature. The parameters
of the gravitative settling term are: Δm the mass difference between two species, g the acceleration due to gravitation, and R the ideal gas constant. The parameter Ω denotes the thermal
diffusion sensitivity, which is specific for each gas and has to be determined experimentally
(Severinghaus et al., 2001). Details are explained in the respective paragraphs.
Processes affecting δ13C of the ice core gas archive
17
2.1.1 Firn structure
The zone on top of an ice sheet is known as firn and is both porous and permeable. For Antarctica its thickness can vary from 50 to 150 m, primarily depending on local climate parameters like mean accumulation rate 8 and mean annual temperature (Kaspers et al., 2004). To be
suitable as an uncompromised gas archive for CO2, surface melting during summer has to be
excluded as CO2 readily dissolves in melt water (Stauffer et al., 2002). Therefore, the following description of the physical processes within the firn column is restricted to cold sites in
central Greenland or Antarctica with dry snow conditions.
Fresh snow has low densities and is highly porous, but is quickly reworked mechanically by
wind and settling. Sublimation processes driven by daily and seasonal temperature cycles further reduce the surface area and snow is transformed to firn (Fig. 2-1). By these processes
Figure 2-1 Snow pit at Kohnen station (75°00’S, 00°04’E, at 2882 m above sea level) showing the upper 60 cm,
roughly four years of the stratigraphic sequence of snow fall or accumulation events. The dashed line marks the
original surface at 08.02.2004. Bright, translucent layers consist of loose material with lower densities than
adjacent layers (marked with bars). These layers represent depth hoar horizons formed by water vapor transport
within the upper 20 cm induced by seasonal temperature gradients. The thin features well pronounced at 11, 24,
39, and 49 cm mark crusts of only a few mm consisting of densely packed snow (marked with an arrow). They
form either due to wind action (wind crust) or when a consolidated surface is exposed to solar radiation between
two deposition events. This small-scale heterogeneity in surface snow properties partially survives during the
firnification process comprising around 800 years at this site. These surface features later modulate the lock-in
process and the depth of the non-diffusive zone and the effects during the bubble close-off.
8
the accumulation rate on a glacier is the net mass balance gain in snow throughout a year. It results from added
snowfall minus loss due to melt or sublimation and horizontal mass transfer due to wind drift. Since the accumulation rate is highly variable between years, the mean of several years is used for characterization.
18
Processes affecting δ13C of the ice core gas archive
distinctive layers form which ultimately determine how and when air is trapped within the ice.
With progressive accumulation of snow layers on top, growing ice crystals are deformed and
by sintering and creep this compaction increases the firn density, in turn the porosity gradually decreases. Ongoing creep finally traps air in bubbles.
2.1.2 Gas transport properties within the firn column
The firn column (Fig. 2-2) can be separated into three zones according to the prevailing type
of air movement (Sowers et al., 1992). First, the so called convective zone, which is advectively well mixed with the overlaying atmosphere due to pressure differences, e.g. wind
pumping (Colbeck, 1989). This zone generally comprises the upper 1 to 10 meters of the firn
column, but might be thicker under special climatic conditions such as in low accumulation
Figure 2-2 Scheme of physical processes occurring in the firn column with depth and age values for ice and gas representative for the drill site Kohnen station. The density increases from ~0.3±0.2 g cm-3 at the top to ~0.8±0.05 g cm-3 at
the firn-ice transition. Ages of the firn air were derived during the field campaign 2005/2006 (K. Weiler and J. Freitag,
personal communication, 2006) by firn air pumping. Gas ages were derived by comparing the measured CO2 concentration profile with the known anthropogenic CO2 rise. The convective zone contains air from the year of sampling due
to rapid mixing. Then gas ages steadily increase with depth due to decreasing effective diffusivities of the porous firn.
In the diffusive zone, gas transport is mainly by molecular diffusion and the gas species and their isotopes fractionate
due to the Earth’s gravitational field. In the region of 80 to 90 m, where most of the bubbles close-off and trap firn air,
the mean gas age is 15 to 20 years. Below this depth the gas ages rapidly increase as prevailing diffusive mixing is
shifted to a unidirectional expulsion of air within the non-diffusive zone. As the age of the ice at this depth is around
850 years (Traufetter et al., 2004), a large Δ-age off-set has to be considered when gas records are compared with
records measured on the ice phase.
Processes affecting δ13C of the ice core gas archive
19
sites or “megadune” regions (Kawamura et al., 2006 and references therein). Large annual
temperature cycles lead to thermal fractionation processes which are only transient and generally do not effect the processes below. The simultaneous occurrence of advective mixing in
conjunction with the observed isotopic gradients due to thermal diffusion do not necessarily
contradict each other because molecular diffusivity of gases in shallow firn is rather fast (1m2
day-1; for details see Schwander et al. (1988)). If convective mixing, i.e. the effective vertical
eddy diffusivity, exceeds this value, isotopic fractionation due to thermal diffusion does not
occur (Severinghaus et al., 2003).
Secondly, the diffusive zone, where the isotopic and elemental composition of the air is altered by molecular diffusion. This zone generally comprises 50 to 100 m depending on the
accumulation rate and the annual temperature of the site. In contrast to the convective zone,
here diffusive processes are persistent over longer time scales (years to decades) as annual
temperature cycles or advective mixing do not reach the diffusion zone. The main driving
forces are gravitative settling and vertical concentration and temperature gradients.
Thirdly, a non diffusive zone of several meters, where unrestricted vertical gas exchange and
diffusion finally cease at the lock-in depth. In contrast, horizontal gas movement remains active along permeable layers. This is due to contrasting differences in vertical firn permeability. Certain layers become nearly impermeable and restrict vertical diffusion, but allow expulsion of air at confined leaks from underlying, still permeable layers. The actual depth and extension of this zone strongly depends on the anisotropy of the seasonal snow and firn layers
and is primarily generated by the accumulation and wind regime of the site. New methodic
attempts to analyze the three dimensional firn structure and to derive effective gas transport
parameters are currently developed, but so far the firn’s anisotropic structure has not been
implemented in models (Freitag et al., 2002).
2.1.3 Enclosure process, age distribution and Δ-age
Figure 2-2 further shows that the bubble enclosure is a continuous process, and bubbles trap
firn air over a wide depth interval starting in the lower diffusive zone. Generally, this happens
when the density of a firn layer exceeds a threshold, the critical density of around 0.83 g cm-3.
This density is reached at different depths for different firn layers according to the seasonal
layering (Freitag et al., 2004). Further, a fraction of bubbles closes-off earlier than this critical density. Consequently, the trapped air in an ice sample has not a discrete age, but can be
described with a characteristic age distribution. For the drill site EDML, Siegenthaler et al.
(2005a) report a mean age distribution of 59±5 years (half height of the distribution) shown in
Figure 2-3. This age distribution acts like a low pass filter and atmospheric fluctuations with
frequencies time scales of years to decades are attenuated in their amplitude (Spahni et al.,
Processes affecting δ13C of the ice core gas archive
20
2003). For atmospheric gases with rapidly changing concentrations, like methane, this leads
to a considerable loss of information at drill sites with low accumulation, hence, broad age
distributions. In contrast atmospheric CO2 changes are generally less rapid, because the large
reservoir of the surface ocean already smoothes out fast CO2 fluctuations.
Due to the firn’s permeability the firn air is in contact with the free atmosphere and therefore
the enclosed air at a certain depth is always younger than the surrounding ice (Fig. 2-2). The
mean age and the width of the age distribution of the firn air increases nonlinearly with depth.
The parameter describing how efficient the vertical gas transport within the firn column can
operate is the effective diffusivity, Deff, introduced in Eqn. 2-1. With increasing depth, firn
density increases (Fig. 2-2) and in turn the remaining pore space volume, or porosity decreases. Not only the volume of the gas phase decreases with depth, but also the shape of the
pore space changes. Pores become more and more elongated and tortuous, therefore molecular gas transport becomes less efficient. Deff accounts for this reduction in pore volume and
increased tortuosity, and can be approximated from the diffusivity of a gas species in free air,
reduced by the available open pore space and tortuosity. Note that since the diffusion velocity
of gases is inversely proportional to the relative molecular mass, different gas species have
different mean ages at a certain depth (Rommelaere et al., 1997). In case of the pair CO2 and
CH4 with masses 44 and 16, the lighter molecule methane diffuses 30% faster than CO2 , accordingly their corresponding mean ages vary at a certain depth (Schwander et al., 1993). For
CO2 at the EDML drill site the mean gas age at the lock-in depth of 88 m was determined to
be about 15 years for the current climatic conditions (Weiler et al., GRL 2006, submitted). At
this depth the age of the ice approaches around 850 years.
Age distribution [yr-1 ]
0.012
0.01
0.008
0.006
width at half heigth
59 ± 5 yrs
0.004
0.002
0
0
50
100
150
200
250
Age [yr]
Figure 2-3 Age distribution for CO2 of enclosed air bubbles at the EDML drill site. For the modeling current
conditions were used as input parameters. The age distribution is normalized to have an area of 1 and the width
at half height was calculated 59±5 years, for details see Siegenthaler et al. (2005a), data provided by Renato
Spahni.
Processes affecting δ13C of the ice core gas archive
21
Therefore, in addition to the age distribution an age difference (Δ-age) between ice and air at a
certain depth has to be calculated. For the EDML core Δ-age is around 825 years for Holocene conditions (Siegenthaler et al., 2005a). The age distribution and Δ-age are strongly dependent on accumulation rate and annual temperature of a site and can vary spatially over an
order of magnitude throughout Antarctica (Kaspers et al., 2004). Knowing the precise Δ-age
is central for deciphering leads and lags concerning changes in temperature and CO2.
2.1.4 Separation of gases and isotopes due to gravitation
In the absence of turbulence, a gas mixture subject to a gravitational field will tend to unmix
by a process known as gravitational settling. The heavier isotope species or gas component
preferentially accumulates at the bottom and light components at the top of a column. A
steady state is reached in which gravitational settling in one direction is balanced by diffusion
along a concentration gradient in the other direction. The isotopic enrichment at the bottom of
the firn column (lock-in depth) can be calculated by applying the barometric equation:
mgz
⎛ ΔRT
⎞
− 1⎟⎟ ⋅ 1000 [‰]
δ = ⎜⎜ e
⎝
⎠
≅
Δmgz
RT
⋅ 1000 [‰]
(2 − 2)
with Δm the absolute mass difference between two gas species, T the mean annual firn temperature, z the diffusive column height (total firn column height minus convective zone and
non diffusive zone). For the conditions currently prevailing at the EDML site (mean annual
temperature = -45 °C, and diffusive column height around 80 m) the enrichment is pretty well
known and is 0.45‰ for Δm = 1 g mol-1 (Landais et al., 2006). Analogue to the isotopic enrichment, gravitative settling results in an increase of the CO2 mixing ratio as well. Assuming
Holocene conditions with 280 ppmv, air at close-off depth is 1.8 ppmv enriched in CO2 compared to the atmosphere (with Δm = 15.2 g mol-1 for the mass difference between CO2 and the
average mass of air).
Climatic conditions during glacial times considerably deviated from the current ones, i.e.
temperature and precipitation were considerably lower and so was the firn structure and the
diffusive column height, and hence the gravitational enrichment with depth.
To overcome the lack of information on the past diffusive column height, δ15N of the trapped
atmospheric N2 can be analyzed and be used to correct for the gravitational fractionation of
CO2. Nitrogen has a mean atmospheric residence time in the order of hundred thousand years
and its isotopic composition can be assumed constant for this approach (Mariotti, 1983). As
shown above, the isotopic fractionation effect of a gas species due to gravitation is a function
of the absolute difference in mass. Hence, the mass difference for 13CO2 with the mass 45 and
22
12
Processes affecting δ13C of the ice core gas archive
CO2 with mass 44 is 1, the same is true for 15N14N with mass 29 and 14N14N with mass 28.
Consequently, δ15N values can be directly used to correct the gravitational effect on δ13C.
According to δ15N measurements from the EDML core, the diffusive column height might
have been reduced during the Last Glacial Maximum (LGM). Isotopic fractionation due to
gravitation was only 0.39‰, instead of the Holocene level of 0.44‰ (Landais et al., 2006).
In addition to this empiric approach, coupled firn densification and diffusion models are used
to calculate the diffusive column height of the firn (Goujon et al., 2003) . According to Eqn.
2-2, additional to the diffusive column height, the mean annual temperature is needed. It can
be derived from δ18O or δD of the ice itself by using the spatial relation of annual temperature
and isotopic composition of the precipitation (Jouzel et al., 2003).
2.1.5 Diffusion along a thermal gradient
A gas mixture within a temperature gradient undergoes thermal fractionation that accumulates
the heavier isotopes preferably at the cooler side, whereas lighter isotopes preferentially accumulate at the warmer side. The same is true for a pair of gas species, e.g. O2/N2. In a constant temperature gradient, a steady state is reached in which thermal diffusion in one direction is balanced by diffusion along a concentration gradient in the other direction. This effect
is called thermal diffusion and is due to the sensitivity of intermolecular forces during collision of molecules and atoms (Severinghaus et al., 2001; Grachev and Severinghaus, 2003 for
methodic details). The isotopic fractionation due to this effect amounts to:
⎡⎛ T ⎞α T
⎤
δ = ⎢⎜⎜ t ⎟⎟ − 1⎥ ⋅ 1000 [‰]
⎢⎣⎝ Tb ⎠
⎥⎦
≅
ΩΔT
(2 − 3)
where ΔT is the temperature difference between the top (Tt) and the bottom (Tb) of the diffusive column, αT is the thermal diffusion constant and Ω the thermal diffusion sensitivity (in
‰ °C -1). The thermal diffusion sensitivity Ω of a gas species is measured as the observed
fractionation divided by an applied temperature difference. Ω and αT are temperature dependent. For some gases Ω was determined experimentally and values range from 0.015‰
°C-1 for δ15N and 0.025‰ °C-1 for δ18O at 255 K and atmospheric pressure (Grachev and
Severinghaus, 2003). Until now, no data was published for the corresponding effect on the
stable carbon isotopes of CO2. However, the values for the other gases suggest, that the magnitude of thermal diffusion has to be critically considered also during the analytical procedure.
In the turbulent atmosphere the effect of thermal diffusion as well as gravitational settling is
not observable since temperature differences induce convective motion of the air, which destroys any fractionation. Only an exceptionally stable atmospheric inversion at a special to-
Processes affecting δ13C of the ice core gas archive
23
pographic setting recently allowed to detect thermal diffusion in the atmosphere (Adachi et
al., 2006). Porous media like extended sand dunes and firn columns are able to suppress convection due to small pore diameter and tortuous pore space geometry. In these cases thermal
diffusion can affect the isotopic composition of the gases involved (Severinghaus et al.,
1996). The required temperature gradient within a firn column results from its thermal inertia
producing temperature differences of up to 20 °C within the upper 5 m (Severinghaus et al.,
2001). Most pronounced are the large seasonal temperature cycles typical for the interior of
polar ice caps, which affect, however, only the upper few meters of the firn column. In contrast, thermal diffusion can sometimes involve the entire firn column. A prominent application
of this effect is the identification and quantification of rapid temperature changes in the
Northern Hemisphere from Greenland ice cores (Severinghaus et al., 1998). As the effect of
thermal diffusion in the firn column requires rapid and large changes (several degrees in annual mean temperature in 1000 years (Caillon et al., 2001)) an influence on δ13C measured on
ice cores is only minor (Kawamura et al., 2006). Similarly to the gravitation effect on δ13C,
the influence of thermal diffusion can be principally corrected with δ15N and ratios of noble
gas isotopes. To do this, gas specific thermal diffusivity constants have to be used and the
effect of gravitation separated from thermal diffusion (Severinghaus et al., 2001).
2.1.6 Diffusion along a concentration gradient
According to Fick’s law, the diffusive flux of a gas species along a concentration gradient can
be described as follows:
dC
(2 − 4)
dx
where F the net flux and dC/dx the concentration gradient in direction of diffusion and D the
F = −D
diffusion constant. The latter is proportional to the inverse of the square root of the molecular
mass. This relation results form the fact that all molecules of a gas mixture have the same
kinetic energy (E=1/2mv²). Consequently, gas species or isotopes with a lower mass move
faster and have higher diffusion constants. If the diffusion involves the movement of one gas
species through a second, here A for relevant gas species of CO2 and B for air, molecular
masses have to be replaced by their reduced masses, denoted with µ:
μ=
m A ⋅ mB
m A + mB
(2 − 5)
In case of the diffusion of CO2 molecules in air (mean mass 29 g mol-1) the isotopic fractionation, 13α, for the stable carbon isotopes of CO2 can be written as follows:
Processes affecting δ13C of the ice core gas archive
24
α=
13
D13 CO
*
2
D12 CO
=
m A + mB
2
*
m A ⋅ mB
m A ⋅ mB
=
m A + mB
45 + 29
45 ⋅ 29
44 ⋅ 29
= 0.9956
44 + 29
(2 − 6)
α the fractionation factor for the pair 13CO2 and 12CO2 with masses 45 (mA*) and 44
(mA), respectively. In delta notation a fractionation factor of 13α = 0.9956 translates into an
with
13
isotopic depletion of 4.4‰.
The two processes described above, gravitative settling and thermal diffusion, create vertical
gradients both in the concentration and in isotopic composition. Further, vertical gradients
within the diffusive firn column occur when the ambient atmospheric composition changes
and the signal slowly propagates downward into the firn column. In either case, ordinary diffusion reduces these gradients until a steady state equilibrium is being established according
to Eqn. 2-1
With the exception of the current anthropogenic CO2 rise, natural changes in the atmospheric
CO2 concentration are marginal in view of the mean age of the firn air. Within the Holocene,
maximum changes were observed during the Little Ice Age reaching 3 ppmv CO2 per 100
years (Siegenthaler et al., 2005a). For the most rapid CO2 increases during the end of the last
ice age, CO2 increase rates of 1-2 ppmv per 100 years were reported (Monnin et al., 2001).
Although these data were derived from Antarctic ice cores from low accumulation sites and
consequently also broad age distributions smoothing the original signal, rapid CO2 fluctuations are unlikely due to the equilibrium with the surface ocean. This gentle behavior of the
atmospheric CO2 concentration due to ‘buffering’ with the large reservoir surface ocean is
known from modeling studies (e.g. Köhler et al., 2005b). Therefore, diffusion effects due to
natural changes of the ambient CO2 on the isotopic composition of CO2 are minor compared
to the gravitational effect. However, for the rapid CO2 rise during the last 50 years this has to
be taken into account. According to Trudinger et al. (1997), the effect caused by the large
CO2 concentration gradient currently found within the firn column is as high as 0.1‰.
2.1.7 Processes during bubble close-off
Contrary to the three processes described above, where diffusion takes place in the gas phase,
the following phenomenon deals with diffusion through the ice lattice. Elemental ratios in air
pumped out of polar firn layers near the firn-ice transition show systematic enrichments of
Ne/N2, O2/N2 and Ar/N2 (Huber et al., 2006; Severinghaus and Battle, 2006). In turn, air extracted from already closed-off bubbles is relatively depleted. This unexpected depletion in
O2/N2 ratios was recognized early, when O2/N2 ratios were measured with the aim to gain
information about the strength of the photosynthetic oxygen production (Bender et al., 1994
Processes affecting δ13C of the ice core gas archive
25
and references therein). The O2/N2 depletion was found to be systematic and cyclical and although the underlying mechanism remained obscure, the fractionation effect was exploited for
dating issues (Bender, 2002). Only recently, the effect of gas fractionation during bubble
close-off gained more attention and can now be described as a special diffusion process
through the ice lattice (Huber et al., 2006; Severinghaus and Battle, 2006). In contrast to gas
phase diffusion, here the mass of the species is not a relevant parameter, but rather its diameter. Large atoms, like Kr and Xe, are not affected by this process, however, the small atoms
He and Ne are subject to considerable losses. Below a critical diameter of 3.6 Å 9, gas molecules and atoms migrate through the lattice and are partially lost to the firn air and are therefore relatively depleted within the bubble. The dependency of the measured loss with diameter is strongly nonlinear (Severinghaus and Battle, 2006).
Analogue to the preferential gas loss during the bubble close-off, a similar diffusion process
through the ice matrix is present during the storage of ice cores after drilling, which causes
further off-sets for the gas composition, e.g. for N2/O2 (Bender, 2002) 10.
CO2 has a molecular diameter of 3.94 Å, thus, should lie on the safe side of this size criterion.
The values for the diameters are collision diameters, derived form viscosity experiments. Note
that besides collision diameters, kinetic sieving diameters are used in the field of studying
diffusion and permeation through porous membranes. Here, the effective diameter of CO2 (3.3
Å) is smaller than for N2 (3.6 Å) and CO2 preferentially passes inorganic membranes (Yeom et
al., 2000 and references therein).
To derive effective diameters for a given process, both the dimension of the gas of interest,
but also for the second collision partner, in our case the ice lattice, have to be considered. The
data basis for effective diameters is sparse, therefore, the conclusion that CO2 is not affected
by this process has yet to be proven. Firn air measurements near the close-off zone did not
indicate any effect on the CO2 concentration. However, in contrast to other gases like O2, N2,
and the noble gases, for which a relative enrichment against a fairly stable background concentration can be detected with relative ease, CO2 is more uncertain due to the anthropogenic
rise. Furthermore, the CO2 concentration profile with depth is often an input parameter to tune
the model, thus, CO2 is not a free parameter, which limits the interpretation.
2.1.8 Clathrate formation and disintegration
Although this process per se does not alter the gas composition within the ice, clathrate formation and afterwards its gradual disintegration after drilling strongly influences the mixing
9
1 Å (Angstrom) = 0.1 nm
owing to diffusive loss of gases during storage, it is common practice for gas analysis to remove the at least 5
mm of the outer layer of an ice sample prior extraction.
10
Processes affecting δ13C of the ice core gas archive
26
ratios of the gas phase. This is crucial since conventional extraction techniques for δ13C in ice
cores are only capable to extract the gas in bubbles. The new sublimation approach described
in this thesis allows to quantitatively extract not only the gas phase, but also the air enclosed
in clathrates. Nevertheless, as changes in the mixing ratio of CO2 are involved with the clathrate disintegration, an isotopic fractionation related to a diffusion processes must be considered.
Figure 2-4 gives an impression of the proportions of bubble ice and clathrate ice for a deep ice
core, exemplified for the EDML core. It is striking that the zone with pure bubble ice, for
which conventional mechanical extraction techniques are aimed for, comprises only the top
500 m or around 15% of the core length. In terms of age, the proportion of bubble ice is even
marginal due to the thinning of the layer thickness with depth. For the EDML core, only the
Holocene is covered within the bubble ice region and >95% of the climate history lies within
the clathrate ice or in the transition zone.
Ordinary ice, Ih, has a hexagonal structure with the density (0.917 g cm-3) considerably lower
than the liquid phase water (1.00 g cm-3) 11. This relatively loose packing of the ice crystal
0
0
firn
bubble ice
8
bubble-clathrate transition
-1000
22
-1500
47
clathrate ice
-2000
80
-2500
176
-45
-40
-35
-30
-25
-20
-15
-10
Age [ka BP]
Drilled depth below surface [m]
-500
-5
Borehole temperature [°C]
Figure 2-4 Schematic sketch of the EDML ice core with respect to the relative proportions of bubble and clathrate ice. The figure further shows the gradual temperature increase with depth as a consequence of the thermal
heat flux from the bedrock. The temperature profile was derived from direct borehole measurements in the field
(F. Wilhelms, unpublished data) For comparison, the ice age for the corresponding depth intervals are shown.
11
both values for atmospheric pressure and 0 °C
Processes affecting δ13C of the ice core gas archive
27
allows to form voids or cages. Clathrates or air-hydrates form, when these cages are occupied
by one or two guest molecules as shown in Figure 2-5. Kuhs et al. (1997) found that these
cages occur in two different sizes, which are occupied by guest molecules of contrasting diameters depending on pressure and temperature. In case of the clathrates found in deep glacier
ice, the guest molecules originate from the enclosed air, therefore, these special clathrates are
often specified as air-hydrates. Nevertheless, the broader term clathrate and clathrate ice is
widely used in the ice core literature.
Air-hydrates have a cubic crystal symmetry and are translucent. This property makes pure
clathrate ice completely transparent in contrast to bubbly ice. To be stable, a set of thermodynamic conditions have to be fulfilled, i.e. temperature, partial pressure of a gas species have to
exceed certain thresholds. For each gas and temperature a special dissociation pressure has to
be reached until clathrates become the stable phase (Ikeda-Fukazawa et al., 2001). The pressure within a bubble gradually increases with depth, with the bubble volume decreasing.
When the dissociation pressure of the clathrate is exceeded, the bubbles gradually disappear
as more and more gas diffuses into the ice matrix to form clathrates. This zone where both
bubbles and clathrates coexist is generally several hundreds meters wide, for EDML approximately from ~500 m to ~1000 m depth and called transition zone 12. Owing to the different
Figure 2-5 The crystal structure of air-hydrate. The red-white frame shows the water host lattice forming two
types of cages in which the oxygen (blue) and nitrogen (green) atoms are located. The resulting structure has
cubic symmetry and belongs to von Stackelberg's type II (unpublished, with permission of W. Kuhs from
http://kristall.uni-mki.gwdg.de/).
12
note, due to the high pressure within the bubbles the ice core is brittle around the depth of 500 m and the ice is
of poor quality and highly cracked. Clathrate formation prevents further pressure built-up and results in good
core quality of the transparent clathrate ice
28
Processes affecting δ13C of the ice core gas archive
dissociation pressures of the air components, the gas composition within a bubble is subject to
changes. Moreover, the atmospheric gas components have different diffusion coefficients
(Salamatin et al., 2001; Ikeda-Fukazawa et al., 2005). In case of the bulk components, O2 and
N2, oxygen diffuses several times faster than nitrogen, hence, the latter becomes progressively
enriched within the bubble. As a result of these effects, large anomalies in the mixing ratios of
several gas components were reported for O2/N2 (Ikeda et al., 1999) as well as for the CO2
concentration within the bubble-clathrate transition zone (Eyer, 2004).
For the gas analysis of ice cores both the bubble-clathrate transition zone and the clathrate
zone pose analytical challenges. Clathrates are metastable and slowly disintegrate due to pressure release after drilling on time scales of years to reform bubbles. Thus, once the gas is
trapped within the clathrate cages, a quantitative extraction with mechanical means is hardly
possible. During the disintegration of clathrates, the mixing ratios of the gases are distorted
again due to kinetics and the gas specific dissociation pressures involved (Ikeda-Fukazawa et
al., 2005).
In contrast to the mixing ratios of gas species, which are largely effected by clathrate formation and disintegration, isotope effects have not yet been reported. This is due to the fact that
the process of filling the clathrate cages is a matter of the geometry of the molecule related to
the available cage size. Consequently, the filling-up of the cages is sensitive to the molecule’s
volume and not to its mass. Though the binding energy of the 13C-16O bond is slightly higher
than for 12C-16O, thus its bond length is shorter, this difference is too small to induce a steric
isotope effect (Mook, 2000). Nevertheless, low extraction efficiencies with mechanical techniques and diffusion effects during the disintegration of clathrates, which are known to cause
isotopic fractionation, call for a quantitative extraction for δ13C (Eyer, 2004).
2.2 Chemical reactions and a possible in-situ production of CO2
In the following, chemical reactions of impurities found in polar ice cores are evaluated for
their relevance for CO2 and δ13C ice core records. 13 Whereas the physical processes reviewed
above strongly depend on climatic parameters of the site (e.g. temperature, accumulation
rate), the chemistry of the ice is primarily influenced by source and transport processes of the
impurities. Accordingly, the geographic position of an ice core drill site with respect to continental source regions and also atmospheric chemistry plays a major role. Chemical reactions
and the issue of in-situ production of CO2 within the ice have been known for a long time
(Delmas, 1993; Anklin et al., 1995; Smith et al., 1997a). Similarly, for the greenhouse gas
13
Organic compounds like acetaldehyde are critical for the mass spectrometric measurement of δ13C on CO2 as
well. Acetaldehyde has a molecular mass of 44 g mol-1, thus it might interfere with the identical masses of CO2 if
it enters the ion source of the mass spectrometer.
Processes affecting δ13C of the ice core gas archive
29
N2O, artifacts were reported as well, especially from core sections with high dust concentrations (Flückiger et al., 2004; Spahni et al., 2005) and additionally for carbon monoxide, CO,
measured in Greenland ice (Haan and Raynaud, 1998).
In addition to atmospheric CO2, ice cores can contain other carbon sources. First, inorganic
carbon, mainly in the form of CaCO3 as a component of terrestrial dust. Secondly, a large
suite of volatile and particulate organic compounds. They originate either from the oxidation
of atmospheric trace gases (like formaldehyde from CH4), are emitted during fire events (organic acids) or transported together within the continental dust particles (eroded soil material).
In total, for Greenland ice these impurities can potentially contribute to an in-situ CO2 production ranging from 140 ppmv within the Holocene to 700 ppmv during the glacial period, assuming these compounds were completely transformed to CO2. Impurity concentrations for
Antarctica are generally lower and the CO2 equivalents amount to 80 ppmv in the Holocene,
while for the glacial period the relevant impurities are more abundant (Tschumi and Stauffer,
2000). Note that for this worst case scenario of the in-situ production Tschumi and Stauffer
(2000) assumed that the carbon containing impurities are stoichiometrically transformed to
CO2. However, an unequivocal contamination of CO2 ice core records was so far shown only
for Greenland, while Antarctic ice is assumed to be unproblematic in this respect.
Beginning with the discrepancies of early CO2 concentration measurements on Greenland ice
cores, the first paragraph reviews the status of in-situ CO2 and its relation to possible chemical
reactions. With the aid of a mass balance approach, the issue of the isotopic composition of a
possible in-situ contaminant is highlighted. The following paragraphs summarize the basic
information concerning possible sources and reactions of organic and inorganic compounds,
which are related to a possible in-situ production. Where available, information about the stable carbon isotopic composition of these compounds is provided to assess their potential effect on the atmospheric δ13C values trapped in the ice core.
2.2.1 Indications for CO2 in-situ production from Greenland ice cores
As shown in the physical section above, molecular diffusion within the firn column smoothes
out any seasonal and short term atmospheric CO2 fluctuations. Nevertheless, CO2 measurements on Greenland ice cores (GISP2, GRIP) conducted with a precision of ca. 3 ppmv revealed variations of 15-50 ppmv within an annual ice layer (Anklin et al., 1995; Tschumi and
Stauffer, 2000). Consequently, the observed CO2 fluctuations cannot be explained with analytical uncertainties, which is confirmed by the analysis of corresponding Antarctic ice cores
showing a scatter one order of magnitude lower (Barnola, 1999; Stauffer et al., 2002; Stauffer
et al., 2003; Siegenthaler et al., 2005a).
30
Processes affecting δ13C of the ice core gas archive
A second argument strongly pointing to a non-atmospheric contribution of CO2 in ice cores is
that also the mean CO2 values from Greenland and Antarctic ice cores do considerably deviate (see Fig. 2-6). Anklin et al. (1995) compared CO2 measurements throughout the younger
Holocene among Greenland and Antarctic ice cores. To rule out systematic effects, the cores
were measured at two different institutes using different measurement techniques. CO2 values
from Greenland cores were found to be considerably higher than the Antarctic values, i.e.
Greenland cores contain so called excess CO2. This difference increased with time leading to
the assumption that the in-situ production is a function of time. Smith et al. (1997a) found
even higher differences in certain parts of the glacial periods (Dansgaard Oeschger events).
The direct comparison of CO2 records from both hemispheres became more precise since dating between different cores was improved after CH4 synchronization of these cores was established (Blunier et al., 1998). It became evident that an interhemispheric CO2 gradient of up to
20 ppmv is in contradiction with the rate of interhemispheric air mass exchange and regional
distribution of CO2 sources and sinks. Currently, the interhemispheric CO2 gradient is ~2
ppmv and mainly the consequence of large emissions in the Northern Hemisphere (Dargaville
et al., 2003). As a consequence, Greenland ice cores must be contaminated in some way.
Large analytical efforts have been made in the following years at the Physics Institute of Bern
to identify the cause of this excess CO2 in Greenland ice (e.g. Tschumi and Stauffer, 2000).
On core sections from different depth intervals highly resolved CO2 records were measured
Figure 2-6 Comparison of CO2 measurements between a Greenland (GRIP) and an Antarctic ice core (South
Pole). The difference between the two records approaches up to 20 ppmv for gas ages of around 1000 years, but
is small for younger ages. This feature could be reproduced from two individual laboratories using different
extraction methods. The hatched area indicates the range, which can be explained by an interhemispheric CO2
gradient. Figure from Anklin et al. (1995).
Processes affecting δ13C of the ice core gas archive
31
and compared with a suite of chemical components analyzed on the same section. These profiles were analyzed with a depth resolution of only a few centimeters. For Greenland ice from
upper core sections this translates to a sub-annual time resolution (Fig. 2-7).
Tschumi and Stauffer (2000) found similarities of the CO2 excess profile with the following
chemical components or parameters: soluble calcium (Ca2+), insoluble dust, hydrogen perox-
Figure 2-7 Highly resolved concentration of CO2 from a Holocene section of the GRIP ice core compared with
continuous records of H2O2, HCHO, Ca2+, and H+ and electrical conductivity of the meltwater as possible sources
of excess CO2. The proton concentration (solid line) was derived from the in-situ dc conductivity of the ice,
whereas the conductivity in µScm-1 was measured after melting the ice in the meltwater (dashed line, left scale).
Concentrations for the chemical compounds and the excess CO2 are given as µmol kg-1 ice. The total CO2 concentration is reported as ppmv. The core section was dated to 2100 years BP (ice age) and comprises around three
annual layers. The missing part is due to a break in the core; after Tschumi and Stauffer (2000). Note that these
strong relations of excess CO2 and chemical impurities were not found in all measured core sections.
32
Processes affecting δ13C of the ice core gas archive
ide (H2O2), formaldehyde (HCHO), electric conductivity of the meltwater, and the directcurrent conductivity measurement (ECM) of the ice. From the ECM the concentration of protons (H+) 14 can be derived as a measure of acidity since the ECM signal is predominantly controlled by the highly mobile H+ ion (Wolff et al., 1997).
With statistical analysis of the profiles, Tschumi and Stauffer (2000) found positive correlations of CO2 excess and chemistry profiles for conductivity of the meltwater (r = +0.6) and
Ca2+ concentration (r = +0.5). In contrast, negative correlations were found for CO2 excess
and the concentration profiles of H2O2 (r = -0.5 to -0.9) and HCHO (r = -0.3 to -0.9). These
findings led to the assumption that the excess CO2 found in these core sections was due to
chemical reactions; either an acid-carbonate reaction or oxidation of organic compounds, or
both at the same time.
Further evidence that the excess CO2 may not be of atmospheric origin came from a few isotopic measurements conducted on suspicious core sections. The δ13C values measured on extracted CO2 samples considerably deviated from the expected atmospheric composition
(around -6.3±0.5‰). On the GISP2 ice core, Smith et al. (1997a) analyzed δ13C on CO2 from
both a LGM section (~17 ka BP) and from the Holocene (~2.5 ka BP). The LGM values were
-4.45‰ and -3.70‰, thus, the isotopic composition of the added carbon source should be isotopically enriched with respect to the atmosphere. This would corroborate the idea that
Greenland ice is contaminated by an acid-carbonate reaction, at least for the dust rich sections
during the glacial period. For the Holocene, however, the δ13C values were -7.23‰ and 7.43‰. In the light of the δ13C data now available from Antarctica (Taylor Dome from Smith
et al. (1999) and Law Dome from Francey et al. (1999)), and assuming no sudden δ13C shifts
during the last 3000 years, these data are surprisingly too negative. Assuming these Holocene
data are not compromised by an analytical artifact like drill fluid contamination (Leuenberger
et al., 1992; Smith et al., 1999), then the excess CO2 should originate from an isotopically
lighter source compared to the LGM. This could be explained by an in-situ contribution from
an isotopically depleted organic compound.
To illustrate the effect of a contamination with different δ13C values, a mass balance calculation was made (Fig. 2-8). To an uncontaminated atmospheric sample with δ13C = -6.50‰ and
280 ppmv an increasing amount of ‘foreign’ CO2 with an individual δ13C value is added.
Lines represent mixing of the two end members. The isotopic mass balance was calculated for
the following δ13C values: 0‰, -11‰, -28‰ and -55‰. These values were chosen to cover
the full isotopic range of possible in-situ reactions from contrasting carbon sources. In this
simple calculation it is assumed that the isotopic composition of the added CO2 from each
source remains constant and does not vary with the added amount. This might happen, when
14
note, in ice core science the terms acidity and proton concentration are often used interchangeably to represent
a measure of the acid content.
Processes affecting δ13C of the ice core gas archive
33
isotopic fractionation during the chemical reaction within a closed system is involved as discussed later (Rayleigh approach). From the mass balance mixing shown in Figure 2-8, it is
clearly visible that already a small amount of added CO2 might alter the atmospheric signal.
The larger the difference between the atmospheric δ13C value and an assumed δ13C value of
the added in-situ CO2, the larger its effect on the mixed signal. As the measurement precision
for the concentration of CO2 with the state of the art ice core methods is in the order of 2
ppmv, small excess CO2 contributions are hardly detectable. In case of a highly depleted contaminant this would only be visible in the δ13C record.
-6.2
original value
13
δ C added: 0‰
-6.4
1 σ uncertainty of measurement
13
δ C added: -11‰
-6.8
-7
-7.2
-7.4
278
1 σ uncertainty of measurement
13
δ C [‰]
-6.6
280
13
δ C added: -28‰
13
δ C added: -55‰
282
284
286
288
290
CO 2 concentration [ppmv]
Figure 2-8 Mass balance calculation to simulate the contamination of different carbon sources on the original
values (280 ppmv and -6.5‰). To this composition foreign carbon was added from 0 to 6 ppmv with contrasting isotopic signatures representing different carbon sources. From top to bottom: 0‰ for carbon originating
from inorganic carbonates, -11‰ for plant or soil carbon from photosynthesis using the C4 pathway, -28‰ for
plant or soil carbon from C3 pathway, and -55‰ for the assumed isotopic value of formaldehyde as a product
from methane oxidation. Additionally shown is the 1 σ measurement precision with 2.5 ppmv for CO2 concentration and 0.06‰ for the isotopic ratio.
2.2.2 In-situ reactions of organic compounds
From the chemicals usually found in the atmosphere only a few groups are actually relevant in
our context and can be sorted out using two criteria. The first criterion is, if the atmospheric
trace compound combines a certain set of physico-chemical properties to be adsorbed on or
deposited with the snow and if the compound is then relatively inert to survive the oxidative
degradation within the surface snow. The second criterion is, whether a compound that made
its way from the atmosphere into ice has additionally the chemical potential to react to CO2.
34
Processes affecting δ13C of the ice core gas archive
Two main groups, or homologous series, were identified in the past and measured in snow
and ice: aldehyds (R-CHO), and mono- and dicarboxylic acids (R-COOH and HOOC-RCOOH) 15. Prominent examples of these homologous series are formaldehyde and acetaldehyde and formic acid, acetic acid and oxalic acid, which were found in snow and ice
(Kawamura et al., 2001; Guimbaud et al., 2002; Houdier et al., 2002; Narukawa et al., 2002;
Perrier et al., 2002).
2.2.2.1 Precursors and sources of organic compounds in ice cores
Characteristic for organic compounds found in snow and ice cores is that only a small amount
is directly emitted, e.g. during fire events (Kawamura et al., 2001), but the main share originates from oxidation of volatile organic precursors within the atmosphere or snow. Terrestrial
and marine ecosystems produce large amounts of volatile organic compounds. To name a few,
methane is predominantly emitted from wetlands and ruminants. Trees emit terpens and marine algae release organic compounds like DMS and alkyl halides (e.g. CH3Cl, CH3Br).
This organic mixture is subject to a suite of photochemical reactions within the troposphere
and stratosphere. The reaction with the hydroxyl radical, *OH, is the most prominent one. As
the reaction from methane to formaldehyde comprises several steps and involves also nitrogen
oxides and O2, Eqn. 2-7 is not stoichiometric.
CH 4 + *OH
hν , O , NO
2
⎯⎯ ⎯
⎯→
HCHO + HO2
( 2 − 7)
Major side products of these photochemical reactions are peroxides, e.g. hydrogen peroxide
and methylhydroperoxide. Note that formaldehyde itself is only an intermediate and reacts to
CO and CO2 (Perrier et al., 2002). In similar reactions also organic acids and many other
compounds are produced and deposited with the snow (Chebbi and Carlier, 1996).
2.2.2.2 Isotopic composition of organic compounds
Only little information is available about the carbon isotopic composition of organic compounds in the polar atmosphere and in snow. Goldstein and Shaw (2003) review the available
information about the carbon isotopic composition of a wide range of organic trace compounds within the atmosphere. To get an overview, it is worthwhile to look into the general
pattern of the isotopic composition of organic compounds. Variability in the δ13C signature of
organic carbon can be found on several levels:
15
note that in organic chemistry R stands for the rest behind the functional group (R may be a carbon chain or
only a H atom in case of the first element of a homologous series)
Processes affecting δ13C of the ice core gas archive
•
35
among major functional plant types in terms of their bulk composition, e.g. plants
using the C3 pathway have mean δ13C values of -28‰, whereas C4 plants values of
around -11‰ (O'Leary, 1981)
•
among the main chemical groups, e.g. lipids are ~10‰ lighter than the bulk organic matter of plants, see reviews by Lichtfouse (2000) and Hobbie et. al., (2004)
•
among the carbon atoms of a single molecule (e.g. the methyl groups of molecules
are generally up to 40‰ isotopically depleted compared to the mean value of the
molecule (Keppler et al., 2004))
Several conclusions can be drawn from above:
First, for a given compound its primary source determines its mean isotopic composition.
Secondly, when a compound was identified to react within the ice and whereby producing insitu CO2, not its mean δ13C value is decisive, but the δ13C value of the carbon atom(s) transferred to CO2 is relevant. This is due the fact that often not the whole molecule is involved in
a reaction, but rather the carbon atom of the most labile group.
An additional effect has to be considered in this context. As most chemical reactions induce
either a kinetic or an equilibrium fractionation, the isotopic composition of the educt and the
product deviate and do not remain constant over time, but change in a predictable way. Generally, the product is isotopically depleted with the heavier isotope and the remaining educt
becomes progressively enriched with the heavy isotope. This is due to the fact that molecules
containing the lighter isotope react slightly faster and binding bonds of the molecules are
stronger for the molecule with the heavier isotope.
A simple approach to illustrate this effect is applying a Rayleigh model to a reaction. For the
situation within an ice sample with a known amount of a reactive compound, an approach
with a given reservoir and one sink or loss process is most applicable (Eqn. 2-8 and 2-9).
δ
=
(
)
⎛ N
1 + δ ⋅ ⎜⎜
0
⎝ N0
ε
⎞
⎟ −1
⎟
⎠
[ ‰]
(2 − 8)
Derived from a mass balance approach (Mook, 2000), where δο denotes the original isotopic
composition of the reservoir prior the reaction, δ the isotopic composition of the remaining
reservoir after the reaction consumed a fraction of it. N0 represents the reservoir size of the
educt prior to the start of the reaction. N/N0 is then the remaining fraction of the reservoir left
after the reaction consumed some educt with a fractionation factor ε. As illustrated in Figure
2-9, the isotopic composition of the remaining fraction becomes successively enriched in the
heavier isotope, hence, the δ values increase (black curve). In turn, the cumulative product
gets depleted accordingly:
Processes affecting δ13C of the ice core gas archive
36
δ cum
=
( 1 + δ0 )
⎛ N
1 − ⎜⎜
⎝ N0
N
1−
N0
1−ε
⎞
⎟
⎟
⎠
−1
[‰]
(2 − 9)
with δcum the average isotopic composition of the reaction product after a certain fraction of
the educt was consumed. In our example, with ε set to 10‰, the relative difference of the instantaneous product (blue curve) to the remaining product is always 10‰. But as the isotopic
composition of the remaining educt (δ ) steadily increases, the absolute value of the product
increases as well. At the end of the reaction, when the educt quantitatively reacted to the
product, δcum approaches the initial value of the educt (δ0 ).
Consequently, to assess the impact of in-situ produced CO2 on the δ13C value of an ice core
sample, the ‘history’ of a reaction is decisive. In other words, if a chemical reaction produces
CO2 from an educt, the remaining fraction of this educt has to be known. If the in-situ reaction
Isotopic composition of educt and product [‰]
30
20
initial value
of the educt
δ 0 = 0‰
10
remaining educt (δ cum)
instantaneous product
0
ε = -10‰
cumulative product (δ)
-10
1
0.8
0.6
0.4
0.2
0
Fraction of the remaining educt = N/N0
Figure 2-9 Rayleigh model describing the isotopic fractionation of educt and product during a reaction proceeding from left (fraction of remaining educt = 1) to right (fraction of educt = 0). In case of an in-situ reaction
the educt is an organic compound and the product the excess CO2. The isotopic composition of the educt at the
beginning of the reaction, δ0 , was set to 0‰. A fractionation factor ε of 10‰ was chosen as the isotopic difference between the educt and the instantaneous product. The black line shows the gradual isotopic enrichment of
the remaining educt as a function of the fraction already removed from a reservoir. The green line describes,
δcum , the cumulative isotopic composition of the reaction product removed from the reservoir; figure adapted
after Mook (2000).
Processes affecting δ13C of the ice core gas archive
37
has already consumed a large fraction of the educt, and this was accompanied with a large
isotopic fractionation, then the measured δ13C of the remaining educt would be relatively enriched. Consequently, not only the fractionation factor ε and the δ13C of the educt has to be
known, but also its remaining fraction N/No.
2.2.2.3 Formaldehyde as an example for in-situ reactions
To asses the extent of a possible impact of in-situ reactions on the δ13C value extracted from
ice cores, knowledge about the isotopic composition of the precursor is helpful. In the following paragraph it is exemplified, how different precursors can have a decisive effect on the
carbon isotopic signature of an in-situ compound.
Formaldehyde as an example for organic compounds was selected since its concentration in
ice cores is relatively high and it may be oxidized to CO2, thus, may contribute to excess CO2
in ice cores. Formaldehyde is a major intermediate oxidation product of methane, but other
hydrocarbons may also react to formaldehyde (Riedel et al., 2005). Atmospheric methane
represents an isotopically highly depleted compound with a pre-industrial δ13C value of
around -50‰ (Sowers et al., 2005). During the reaction with OH radicals, a fractionation of
~5‰ occurs, thus, the reaction product formaldehyde should approach a δ13C value of around
-55‰, if methane were its main source (Brenninkmeijer et al., 2003). When formaldehyde is
oxidized within the ice by H2O2 (Eqn. 2-10), the formed CO2 is likely to be isotopically depleted as well. This in-situ CO2 increases not only the CO2 concentration, yet the effect is
even worse when considering its impact on the isotopic composition of the atmospheric CO2
in the ice. Note that neither the pathway nor the potential location of this chemical reaction
within the ice were already identified. Interestingly, for some chemical compounds it was
found that concentrations are not equally distributed within the ice, but enriched at the grain
boundaries and especially in triple junctions 16, as revealed from microanalysis (Barnes et al.,
2003; Ohno et al., 2005). It was speculated that even at temperatures as cold as -20 °C a quasi
liquid layer might exist due to the lowering of the melting point caused by a local enrichment
of chemical impurities. As a consequence, this layer may provide a liquid compartment for
chemical reactions (Weller et al., 2004; Ohno et al., 2005).
2 H 2O2 + HCHO →
3H 2O + CO2
(2 − 10)
The partner of formaldehyde, H2O2, is mainly generated during photochemical reactions when
organic compounds react with OH radicals (Frey et al., 2005) and several studies report H2O2
measured in Antarctic snow and firn samples (Gillett et al., 2000; Hutterli et al., 2003).
16
triple junctions or veins are located where three crystal boundaries meet
38
Processes affecting δ13C of the ice core gas archive
Recent studies modeling formaldehyde concentrations within the Antarctic firn and in the
troposphere showed that other sources in addition to methane are needed to explain the measured formaldehyde concentration (Riedel et al., 2005). Other volatile precursors like ethane
and dimethylsulfide may contribute to the formaldehyde budged. The few existing δ13C
measurements on atmospheric formaldehyde are less depleted than one would assume if
methane is its major source. Goldstein and Shaw (2003) report δ13C values for formaldehyde
in remote areas from -17 to -23‰. Similar high values are given in Brenninkmeijer et al.
(2003), but they interpreted the unexpectedly high δ13C values of formaldehyde as a result of
a large kinetic fractionation occurring during a loss process of formaldehyde.
2.2.3 In-situ reactions of inorganic carbonate
Compared to the in-situ issue caused by organic compounds, where many species are potentially involved and reactions are unclear, the acid-carbonate reaction is straight forward. It is
in principle one single carbon species, carbonate CO32-, which reacts with protons 17 to form
CO2 in a two stage acid-base reaction. Carbonate occurring in dust particles is most likely
associated with calcium or magnesium and nearly insoluble in pure water. Hence, its dissolved concentration is low compared to the solid phase.
CO3 + H +
2−
↔
HCO3
−
↔
CO2 + H 2O
HCO3 + H +
−
(2 − 11a )
(2 − 11b)
As shown in the reaction Equations 2-11a and b, in total two moles of protons are consumed
to produce one mole CO2. Further, both reactions are fully reversible and proceed until an
equilibrium state is reached. Hence, high CO32- and HCO3- levels can actually shift the equilibrium, then CO2 is ‘consumed’ from the gas phase and solid phase carbonate being dissolved similar to the weathering process of limestone. This might be the case for carbonaterich sections of Greenland ice cores. Here, partly lower CO2 values than expected were occasionally measured (Smith et al., 1997b). The possibility to reach a chemical equilibrium is in
contrast to the oxidation reactions of the organic compounds discussed before, which have
higher activation energies, thus, the back reaction is hardly possible.
Similar to the reaction of organic compounds, a local liquid phase along the grain boundaries
and triple junctions is assumed as a reaction medium at the temperatures prevailing in polar
ice sheets. Especially the acidic compounds sulfuric acid and methanesulfonate are preferentially located along the crystal boundaries (Barnes et al., 2003). Further, different types of salt
inclusions were found in Antarctic ice cores, which might indicate that carbonate particles
17
for simplicity H+ is used here instead of H3O+ used in formal aqueous chemistry
Processes affecting δ13C of the ice core gas archive
39
were already neutralized during the atmospheric transport by the acid-carbonate reaction
(Ohno et al., 2006). The authors found undissolved CaSO4 particles in ice samples of the glacial period, which contains higher proportions of continental dust than during the Holocene.
CaSO4 is produced when CaCO3 reacts with sulfuric acid, whereby CO2 is liberated. Yet, it is
not fully certain from these findings when the reaction of carbonate with the acid occurred,
either during transport, during the snow and firn metamorphosis prior to the bubble close-off,
or after bubble close-off. Only in the last case, CO2 is trapped within the ice and can contaminate the atmospheric CO2 composition.
2.2.3.1 Sources of carbonate in ice cores and isotopic composition
The main geographic sources of carbonate particles transported to the ice sheets are the dry
regions of the continents. Only a negligible fraction stems from marine sea spray aerosols
emitted from the open ocean due to wind action. With respect to their formation and composition, carbonate particles can be separated into three groups: First, from mechanical crumbling
of rocks due to the action of glaciers. Often marine limestone was ground to fine particles
blown out from the glacial foreland and river banks. This type is classic for the European
loess deposits during the ice ages. δ13C values of marine carbonates are close to the mean isotopic composition of the ocean (around 0‰). Secondly, limnic carbonates are formed by
chemical weathering of soil and rocks and the subsequent transport of these solutes to lake
basins. Dissolved inorganic carbon in rivers and carbonates from lake deposits have a higher
variability in δ13C and are sometimes isotopically lighter compared to marine carbonates
(Mook, 2000; Brunet et al., 2005). Lakes may dry out in phases of reduced precipitation
and/or evaporation and lake sediments may be exposed and the be blown out. A special case
are the dry regions of Taklimakan Desert, China, which is assumed to have provided the high
dust and carbonate levels in the glacial ice of Greenland. Here, δ13C values of around 0‰
were measured in surface carbonates, which was explained that these carbonates are mainly of
marine origin with its isotopic composition nearly unaltered (Wang et al., 2005). Thirdly, pedogenic carbonates mainly originate in semiarid soils and can be blown out after surface erosion by wind action. The δ13C values of pedogenic carbonates are influenced from the vegetation type and other parameters, but are generally within the range from -10 to 0‰ (Cerling et
al., 1989; Ding and Yang, 2000; Stevenson et al., 2005). Owing to the different genesis and
geographic origin of carbonate aerosols, the ratio of CaCO3 on the total mineral dust may
considerably vary as well its carbon isotopic composition. According to Equations 2-11a and
b, the liberation of CO2 from carbonate particles is dependent on the availability of acidic
compounds within the ice. The mere presence of carbonate in ice does not result in any CO2
in-situ production. Smith et al. (1997a) reported that the highest in-situ production was found
40
Processes affecting δ13C of the ice core gas archive
in ice samples originating from the warmer interstadials, where dust levels and hence the carbonate input was reduced, yet the acidic compounds were higher. In these core sections carbonate and acidic aerosols occurred simultaneously without reacting prior to the bubble closeoff. The CO2 concentration of the cold stadials characterized by the highest dust and calcium
levels shows only little off-sets to the Antarctic values. As the acidity concentration derived
from the ECM is zero, the conclusion can be drawn that this was already the case during the
deposition of the snow.
2.3 Conclusions and requirements for the δ13C analysis of ice cores
From the considerations about physical and chemical processes discussed above, the general
conclusion can be drawn that ice core δ13C values can considerably deviate from the original
atmospheric value. Both physical and chemical processes were found to complicate and challenge the ultimate goal of δ13C analysis of ice cores, i.e. to provide reliable data to reconstruct
the global carbon cycle of the past. However, for the dominant fractionation process, the
gravitative settling in the firn column, a sound physical theory can describe this phenomenon,
and with the δ15N value a measurable parameter is available to correct for this effect.
The other effects are still less understood and have to be critically evaluated, if and under
which circumstances they actually disturb the atmospheric signal trapped within the ice. Especially for the chemical in-situ production of excess CO2 only little is known which organic
precursors are oxidized to CO2 and how this happens. Most studies about chemical in-situ
reactions were conducted on Greenland ice cores, which are now considered inadequate for
accurate CO2 and δ13C measurements. Generally, Antarctic ice cores contain considerably less
impurities and the mean annual temperatures are lower, thus, reaction rates are substantially
reduced. Consequently, the small effects, which might also occur for Antarctic ice, are more
difficult to detect with the current detection limits of the used analytic methods. Analyses on
Antarctic ice cores, so far, gave no clear hint that they do not accurately represent the original
atmospheric CO2 concentration. However, highly resolved CO2 concentration measurements
on the two EPICA ice cores (EDML and EDC) revealed an unexplained scatter especially in
young Holocene ice of EDML, which is beyond the analytical precision (Stauffer et al., 2003;
Siegenthaler et al., 2005a). But, the authors state that the scatter of the samples is only
slightly higher than the scatter for standards and analytical off-sets cannot be ruled out.
New analytic developments are underway at the Physics Institute of Bern (Dissertation U.
Federer) to measure the total amount of organic compounds at a high depth resolution. Until
then, the most pragmatic approach followed in this study for δ13C analysis is first to carefully
analyze ice core sections on the respective scale where a suspicious process might induce a
δ13C variability to be detected.
Processes affecting δ13C of the ice core gas archive
41
2.3.1 δ13C artifacts as a matter of scale
The scale of those processes able to induce an off-set in the gas composition between ice
cores and the atmospheric value, can vary from a few centimeters to several hundred meters.
In terms of time this translates to time scales from sub-annual to centuries/millennia for typical conditions occurring in low accumulation regimes. In the following, the physical and
chemical processes are divided according to the scale they affect. Not all listed effects have a
direct impact on CO2 or δ13C, but in case of the Δ–age difference of gas to ice, the effect
might be relevant for comparing and dating different ice cores. The principal aim is to illustrate at which scale these effects might add unexplained variability, to decide at which scale
the analytic has to look at.
effects on the cm-scale may result from:
•
chemical in-situ reactions since concentration profiles of impurities strongly vary
among deposition events and especially on seasonal time scales
•
size dependent fractionation during the process of bubble close-off as found for major
atmospheric constituents (Huber et al., 2006; Severinghaus and Battle, 2006)
•
post coring diffusion processes during the years of storage induce gas fractionation in
the outermost mm to cm of the ice samples (Craig et al., 1988; Bender et al., 1995;
Ikeda-Fukazawa et al., 2005)
•
inhomogeneous clathrate formation/destruction within the bubble-clathrate transition
zone influences the mixing ratio of gases and causes varying extraction efficiencies
with mechanical techniques
effects on scales of up to several 100 m may result from:
•
altered thickness of the firn’s diffusive column height and rapid temperature fluctuations induce changes both in the extent of the fractionation processes, e.g. gravitative
settling, but also for parameters like Δ−age and the age distribution
•
chemical in-situ reactions I: due to principal changes in the mean concentration of
impurities caused by major climatic changes (glacial-interglacial)
•
chemical in-situ reactions II: due to the time dependent reaction rate the in-situ effect
might gradually increase with time and depth as found in the growing difference between Greenland and Antarctic ice cores for older core sections (Anklin et al., 1995)
•
chemical in-situ reactions III: due to increased reaction rate with increasing temperature when ice becomes warmer owing to the geothermal heat flux from the bedrock
Processes affecting δ13C of the ice core gas archive
42
2.3.2 Analytical requirements
The analytical procedure must be capable to precisely analyze δ13C and CO2 on the respective
scale and should meet the following analytical criteria for:
(1) sample size
The sample size should be small enough to detect an artifact on the respective scale,
i.e. a few cm for chemical variations. Further, as ice is generally limited, the method
should allow to make replicates and highly resolved profiles on a single core section.
(2) physical state of the trapped gas
The method should work indiscriminately for bubble and clathrate ice, i.e. the efficiency of the extraction must not depend on properties of the ice itself. This was not
possible until now with the mechanical extraction devices, which extract trapped air
only to 40-80% depending on the type of ice and storage history. Sublimation under
vacuum allows an extraction efficiency of 100% and this technique was already used
to measure CO2 concentration on ice cores, but not for δ13C.
(3) isobaric interferences during the measurement
The mass spectrometric measurement of δ13C should not be compromised by isobaric
interferences. Several organic compounds, like chloro-fluoro carbons used as densifier
for deep drilling, produce mass fragments during the ionization in the mass spectrometer, which are identical to CO2. The state of the art analytical systems for δ13C ice core
measurements were not able to separate these compounds from CO2.
(4) precision
The analytical error of the method should allow to detect possible artifacts in δ13C in
the order of 0.05‰. This precision is necessary as the entire variability of atmospheric
δ13C on glacial/interglacial time scales does not exceed the range of 0.5-1‰.
3 Methods and instruments for δ13C and CO2 analysis on ice cores
3.1 Introduction
The challenge to explain the glacial/interglacial variations in atmospheric CO2 has been a task
of outstanding importance for the paleoclimate community during the last decades and still is
(Indermühle et al., 1999; Brovkin et al., 2002; Broecker and Clark, 2003; Köhler and Fischer,
2004). Although CO2 data are now available for up to six glacial cycles with different resolution (Fischer et al., 1999; Petit et al., 1999; Monnin et al., 2001; Siegenthaler et al., 2005b), a
long, highly resolved and trustworthy record of δ13C is still awaited.
Chapter 3.2 summarizes the technical approaches developed during the last 25 years to analyze δ13C on ice cores. A technical evaluation of the sublimation methods previously used for
CO2 extraction provides Chapter 3.3. The focus is put on the limitations and methodological
shortcomings encountered during their usage to derive the essential requirements to improve
these techniques. The general layout and concept of the newly developed methods in this
work is presented in Chapter 3.4. In Chapter 3.5 and 3.6 the technical details of the sublimation extraction system followed by the tube cracker system are presented. The actual sample
preparation and measurement protocol is described in Chapter 3.7. The data processing and
the procedure verification with blanks and reference gas standards and the necessary corrections are discussed in Chapter 3.8 and 3.9.
3.2 Previous approaches for δ13C on ice cores using mechanical devices
Starting in the early 1980s, first analytical attempts were initiated to extract CO2 from ice
cores to measure its isotopic composition (Tab. 3-1). Generally, the entire procedure comprises three steps: (1) release of the gases occluded in the ice, i.e. separation of the gas phase
from the ice matrix, (2) followed by a sample clean-up to remove traces of water and to separate CO2 from the bulk air 18 constituents, (3) finally the measurement of the isotopic composition in an isotope ratio mass spectrometer (IRMS) using two different inlet techniques: (1)
dual-inlet micro volume and (2) continuous flow (CF). For an overview of the presently used
IRMS approaches see review from Ghosh and Brand (2003).
Although the size of the sample needed per analysis greatly varied between ~5 g and 1 kg, all
extraction techniques were mechanical, only differing in the way the ice was crushed. Typically, an ice sample was loaded into a stainless steel vessel, atmospheric air was pumped off
using vacuum, and then the ice was crushed using either a needle cracker, milling device, or a
18
N2, O2, Ar and other atmospheric compounds which are not condensable at liquid nitrogen temperature (-196
°C); in any cryogenic application CO2 and N2O are trapped together due to nearly identical physical properties.
Methods and instruments for δ13C and CO2 analysis on ice cores
44
cheese-grater. The sample clean-up and transfer, however, considerably varied according to
the applied inlet coupling for the IRMS (Tab. 3-1).
(1) For the dual-inlet micro volume technique, the extracted air was purified in large diameter
vacuum lines whereby either all air constituents were condensed together on a cold head
at 15 K and CO2 is separated afterwards from the air (Friedli et al., 1984; Friedli et al.,
1986; Siegenthaler et al., 1988). Or, CO2 was immediately frozen out in cold traps held at
liquid nitrogen temperature (LN at -196 °C), with the remnant air then collected on a cold
head (Leuenberger et al., 1992). A similar approach was used by Smith et al. (1997a),
Indermühle et al. (1999) and Fischer et al. (2003), with the difference that CO2 was
separated from H2O in two consecutive steps. Afterwards, CO2 was transferred in a glass
tube and stored for the later IRMS measurement. The most elaborate set-up was used by
Francey et al. (1999), who condensed the whole extracted air first on a cold head, and in a
Table 3-1 Compilation of methods for δ13C analysis on ice cores developed from several groups. The applied approaches differ widely in (1) required amount of ice per sample, (2) the extraction technique to liberate the gases
from the ice and (3) the attained extraction efficiency, (4) the sample clean-up and the preparation of the gas sample
prior to its measurement and (5) the type of inlet linking the mass spectrometer. The sample size has been dramatically reduced since the advance of a new inlet coupling, i.e. from dual-inlet micro volume (DI) to continuous flow
(CF) isotope-ratio mass spectrometry (IRMS). Extraction efficiencies for bubble ice were reported to be ~80%, but
lower for clathrate ice (values in brackets). * denotes interferences with drill fluid; ** denotes interferences with
organic solvents (ethanol and isopropanol)
sample
size [g]
crushing
device
recovery
[%]
Friedli et al., 1984
Friedli et al., 1986
Siegenthaler et al., 1988
~700
milling
cutter
80-90
complete trapping of released air and
CO2 on cold head (14 K), then cryogenic
separation of CO2 and air (offline)
DI
Leuenberger et al. 1992
~500
milling
cutter
75-85
cryogenic CO2 trapping prior the air is
condensed on cold head (20 K), *
DI
Francey et al. 1999
1000
‘cheese
grater’
N/A
complete trapping on cold head (15K),
then CO2 extraction under controlled
pressure for viscous flow (offline), **
DI
Smith et al., 1997a
Indermühle et al., 1999
Fischer et al., 2003
200
rotary
crusher
N/A
cryogenic separation of CO2 from air,
flame-sealed glass ampoules (off-line), *
DI
Leuenberger et al., 2003
Eyer 2004
5-10
needle
cracker
75 (50)
dried air expanded in large volume, separation of CO2 from air in a continuous He
flow at 3 bar using the Precon system
(online), *, **
CF
authors
clean-up and transfer to IRMS
inlet
Methods and instruments for δ13C and CO2 analysis on ice cores
45
second step they trapped CO2 off the gas stream using a pressure control unit. Using this
technique, viscous flow conditions within the LN traps of the vacuum line were maintained during the entire trapping procedure. Nevertheless, vacuum lines used for the sample clean-up prior to dual-inlet IRMS measurement are not suitable for more advanced
sample clean-up techniques like gas chromatography (GC). Most studies reported problems during the IRMS measurement caused by isobaric interferences from organic impurities and N2O. The organic impurities either originate from drill fluid contamination or
were introduced during the sample preparation within the laboratory (organic solvents
used in cold traps and cooling systems). Additionally, ice cores naturally contain traces of
organic compounds like acetaldehyde, which may cause isobaric interferences during the
IRMS measurement. The dual-inlet technique has the additional drawback that large sample amounts were needed (up to 1 kg) since only a small fraction (<1%) of the sample actually enters the ion source of the IRMS. However, the attainable precision of the dualinlet technique is unrivaled and can be as low as 0.01‰ for atmospheric measurements on
flasks samples (Ciais et al., 1995; Werner et al., 2001). Yet, the outstanding measurement
performance of dual-inlet cannot be exploited as the overall precision of the whole procedure is determined and deteriorated by the sample preparation and from artifacts.
(2) In contrast, the continuous flow (CF) technique, which became available during the
1990s, principally allows to separate CO2 from interferences using a GC, since the extracted gases are introduced into a continuous He flow. In a typical CF application, CO2 is
transferred and purified within small diameter capillaries using a pressurized He carrier.
The sample is then transferred to the ion source of the IRMS using a so called open split
device. Contrary to dual-inlet applications, here typically 30% of the sample enters the
IRMS. Consequently, CF applications reduced the sample size for the analysis of up to
two orders of magnitude. Leuenberger et al. (2003) were the first employing this elegant
sample preparation technique for δ13C analysis on ice cores by connecting the established
needle cracker (Monnin et al., 2001) to a new developed CF preparation system. Here, the
released air from the extraction device expanded into a large expansion volume. Introduced in a He flow, CO2 was then transferred to the CF system, where CO2 was cryofocused and then admitted to the IRMS. Although principally possible with this CF system,
Leuenberger et al. (2003) did not implement a GC separation. With this approach they
dramatically reduced the sample size down to 5-10 g ice, which allowed highly resolved
δ13C measurements for the first time.
This CF approach was applied in the work of Eyer (2004), who measured δ13C at high resolution on both EPICA ice cores (EDML and EDC). Though the same age intervals for both
cores were measured under the identical preparation protocol, the δ13C values between the
46
Methods and instruments for δ13C and CO2 analysis on ice cores
two cores differed substantially and showed a large scatter (Eyer, 2004). This unexpectedly
large scatter cannot be explained by an atmospheric signal as the variance was found on a
sub-annual scale, which should be smoothed out in ice cores due to the slow bubble enclosure
process. As the δ13C results depended on the measured ice core and the precision of the reference gas measurements was 0.12‰, Eyer (2004)speculated that the scatter in EDML might be
due to a process connected with the ice. Suggested candidates were fractionation during the
bubble close-off and in-situ reactions, which are both relevant at the scale of a few cm. Since
the sample size used in former studies was large (up to 1 kg, see Tab. 3-1), thus, small scale
effects might have been unresolved, the results of Eyer (2004) challenged the δ13C approach.
On the other hand, inherent limitations and problems with the ice core analysis itself might
explain the large scatter as well. Eyer (2004) reported sporadic outliers with unusual negative
δ13C values, which he attributed to drill fluid contamination. These isobaric interferences during the IRMS measurement were reported in earlier studies as well (see references in Tab. 31). It can be concluded that an effective sample clean-up using a GC helps eliminate this issue. Chapter 3.6 describes the CF-IRMS system developed in this study, where a GC sample
clean-up is applied for the first time for the δ13C analysis on ice cores.
A second critical issue reported by Eyer (2004) was associated with the analysis of clathrate
ice and confirmed also by Siegenthaler (2006), who found out that the extraction efficiency
for clathrate ice (1) is considerably lower than for bubble ice and (2) both the efficiency and
the CO2 mixing ratio was strongly dependent on extraction time. Both studies came to the
conclusion that the nonquantitative extraction of clathrate ice using mechanical devices fractionates at least the gas species (elemental fractionation). Eyer (2004) additionally found also
strong effects on the carbon isotopic composition with more negative δ13C values for deeper
ice with higher clathrate content. Chapter 3.5 describes the developed sublimation system,
which allows a quantitative sample extraction both for bubble and clathrate ice. Until now,
sublimation of ice was only applied for concentration measurements and is used here for the
first time for δ13C on ice cores.
3.3 Quantitative extraction techniques for ice cores – sublimation in vacuum
Having in mind the low efficiency of mechanical extraction devices for clathrate ice together
with elemental and isotope effects reported by Eyer (2004), the urgent need for a quantitative
gas extraction for the analysis of δ13C on clathrate ice is apparent. Note that pure bubble ice
comprises only ~20% of the total length of a deep ice core from Antarctica, in terms of climate history the fraction is even smaller. The overwhelming part of the atmosphere’s climate
history is found in clathrate ice as pointed out in Chapter 2.1.
Methods and instruments for δ13C and CO2 analysis on ice cores
47
For the measurement of the concentration and isotopic ratios of most gases from ice cores
(e.g. N2, N2O and δ15N(N2O), CH4 and δ13C(CH4)) a melt extraction is used (Chappellaz et
al., 1997; Flückiger et al., 1999; Sowers et al., 2005; Bernard et al., 2006). With restrictions,
also the CO2 concentration can be analyzed using a wet extraction for Antarctic cores with
low impurity concentrations, but results deviated from conventional data in certain core sections (Kawamura et al., 2003). A δ13C analysis using a wet extraction technique is probably
not possible. This is primarily due to chemistry as a liquid phase provides favorable reaction
conditions for the impurities (acid-carbonate, organic compounds and H2O2) as discussed also
for the in-situ effects (Tschumi and Stauffer, 2000; Kawamura et al., 2003). Since the δ13C of
calcite or organics substantially deviates from the atmospheric value, as shown in the Chapter
2.2, already a 1% contamination (e.g. 2 ppmv to 200 ppmv CO2) could shift the δ13C up to
0.2‰. A second difficulty is the high solubility of CO2 connected with its isotopic fractionation during gas-liquid transfer from the aqueous HCO3-/H2CO3 system (Anklin et al., 1995;
Zhang et al., 1995).
Thus, the only extraction technique which destroys the ice matrix and enables 100% extraction efficiency and avoids the above mentioned complications is sublimation under vacuum.
Several sublimation techniques to quantitatively liberate air from polar ice cores were developed during the last 15 years (Tab. 3-2). First attempts to sublimate ice in vacuum were conducted by Wilson and Donahue (1990; 1992) with the aim to measure 14C on ice samples to
establish a means for absolute dating of ice cores. Sample sizes as large as several kg were
processed in vacuum lines. The ice was sublimated in a glass vessel using infra red (IR)
lamps, water vapor was removed in large external water traps cooled with organic solvents at
-80 °C, and CO2 was separated from air in LN traps. Interestingly, the authors used a molesieve to collect the air content, mainly to be independent from external vacuum during almost
18 h of sublimation. As the aim was to carbon-date the extracted CO2, it was essential to assure that during the sublimation no ‘dead carbon’ from the CaCO3 rich Greenland ice would
Table 3-2 Compilation of sublimation methods for quantitative extraction of CO2 from ice cores. Note that all
approaches used organic solvents for the H2O cold traps (typically ethanol at -90 °C).
sample
size
[g]
sublimation
temperature
[°C]
energy source
collection of
CO2 and air
3000
-10
infra red lamps
LN cold trap and
molesieve at LN
Güllük et al., 1998
20-50
-36 to -27
infra red lamps
cold head (18 K)
Siegenthaler 2002
15
-36
CO2 laser
cold head (18 K)
author
Wilson and Donahue, 1990
Wilson and Donahue, 1992
Wilson and Long, 1997
48
Methods and instruments for δ13C and CO2 analysis on ice cores
release CO2 due to an acid-carbonate reaction. Consequently, to rule out surface melting, the
sublimation temperature was held below -10 °C and controlled by the resulting vapor pressure
over ice (~2.5 mbar). In a later version (Wilson and Long, 1997), the sublimation apparatus
was coupled on-line to the dual-inlet of a IRMS to allow also measuring δ13C on ice core
samples, however, without having success.
Güllük et al. (1998) built a new sublimation apparatus for the simultaneous concentration
measurement of several greenhouse gases (CO2, N2O, CH4) on polar ice cores. This set-up
was considerably smaller and designed for a maximum sample size of 50 g ice. In contrast to
the approaches of Wilson and coworkers, who had no possibility to admit reference gas to the
apparatus, Güllük’s approach allowed to manually introduce air standards to the sublimation
vessel to check for artifacts and blanks. Instead of using LN cold traps and a molesieve to
separate CO2 from air, a cold head at 15 K was used to condense the entire air sample. As the
intention of this set-up was to measure the CO2 concentration with infra red laser spectrometry, separation of CO2 from the air was not necessary. Unfortunately, the system suffered from
large blank values, attributed to outgassing from the internal surface of the large water traps
and tubing of the all-glass vacuum line. For CO2, blank values were as high as 3-10% of the
sample concentration, i.e. 10-25 ppmv. The authors found a second effect, which runs contrary to outgassing. When an air standard with known concentration was admitted to the dry
sublimation vessel, meaning without adding ice, large loss effects of up to one third of the
original concentration were observed. This interplay of adsorption on and desorption from the
surfaces of the apparatus was related to the partial pressure of water (pH2O) as well specific to
the material as already reported by Zumbrunn et al. (1982). Further, outgassing of CO2 increased when surfaces were unintentionally heated due to reflections of the IR-lamps used for
sublimation. Problematic was also the extensive usage of Viton gaskets joining the sublimation vessel and the water trap.
To improve on the shortcomings of this approach, Monnin (2000) and Siegenthaler (2002)
modified this system: First an IR laser replaced the IR lamps to prevent unintentional heating
of the glass apparatus. A special optical window was inserted to transmit the laser radiation
(~10 µm) to the ice. With the laser set-up, only a part of the ice cube can be sublimated due to
the geometry of the optical window and lenses. Hence, the ratio of sample CO2 to the surface
area of the apparatus decreased. Secondly, the system’s internal volume and area was reduced
using smaller cold traps, but still high blank values persisted and the observed scatter in the
ice core results was higher than for the mechanical extraction (Siegenthaler et al., 2005a).
The two main general problems of this approach, the large surface area of the all-glass apparatus and the extensive usage of polymer material for seals and gaskets could not be reduced
further within this technical set-up.
Methods and instruments for δ13C and CO2 analysis on ice cores
49
All approaches described above used organic solvents for the cooling traps. For the analysis
of CO2 concentrations this causes no problems. However, several studies measuring δ13C on
ice cores reported that minute amounts of organics caused isobaric interference during the
IRMS measurement (see references in Tab. 3-1). These authors recommended to eliminate
volatile organic solvents during the sample preparation, but also to keep the air in the laboratory of the mass spectrometer free from solvents.
To overcome the technical limitations of the previous sublimation approaches and to meet
also the additional requirements for the δ13C measurement, a new sublimation apparatus was
designed. The approach of this work profits from three technical novelties.
•
The system was constructed to minimize the surface effects. The sublimation vessel and
the water trap are merged into one compact vessel, thus, large cross section tubing and
seals to connect these devices are unnecessary. Only all-metal seals and valves are used
and diameters of the metal tubing for the LN cold traps are kept as small as possible.
Thus, the surface area of the apparatus was considerably reduced compared to previous
approaches. This system is described in Chapter 3.5.4.
•
Volatile solvents are excluded from any analytical step. To achieve this, a cooling system
for the combined watertrap-sublimation vessel based on cold air replaced the solvents in
the cooling baths of the previous studies. Further, a special water trap cooled with liquid
nitrogen has been developed. These systems are described in Chapter 3.5.3 and 3.5.5.
•
A reference inlet was constructed to introduce whole air standard to the sublimation vessel
at viscous flow conditions. This device mimics the continuous release of air during the
sublimation and aims at an identical treatment of sample and standard; the reference system is described in Chapter 3.5.8.
3.4 General layout of the entire method
This paragraph introduces the general strategy and set-up of the developed analytical method.
Similar to the previous systems constructed to analyze ice cores for δ13C, the entire analysis
consists of three principal steps:
1. liberation of the trapped air from the ice
2. separation and purification of CO2
3. measurement of the pure CO2 in the IRMS
These three steps can be either coupled on-line, whereby the sample follows the whole chain
of the analytical processes without interruption until the measurement is completed; or alternatively, using an off-line coupling, for which the chain of processes is interrupted at a certain
50
Methods and instruments for δ13C and CO2 analysis on ice cores
step and the follow-up steps are decoupled from the first steps. In the field of stable isotope
analysis, off-line coupling can be advantageous where one step is especially time consuming,
whereas the other steps can be processed more quickly. Since the gas extraction from the ice
is by far more time consuming than the actual IRMS measurement, off-line coupling was chosen and the entire process was split into two separated preparation lines. The two lines fulfill
the following tasks:
•
The first line liberates the trapped air from the ice sample and separates CO2 from the air
in a vacuum line. At the end, CO2 is transferred into a tiny glass tube for storage.
•
The second line purifies CO2 from N2O and other isobaric impurities in a continuous
flow system leading to the IRMS measurement device (CF-IRMS).
This approach enables to measure a large number of extracted ice samples with the CF-IRMS
system within a short time span and, thus, takes benefit from identical measurement conditions with the IRMS for a large set of samples. This is crucial as changes both in the performance of the IRMS, but also in the CF system are well known problems. Following this technical partition, first the setup of the ‘sublimation extraction’ is presented and then the CF system called ‘tube cracker-GC’ system leading to the IRMS (Fig. 3-1 and 3-2). With the sublimation extraction (Fig. 3-1), an ice sample is continuously sublimated and CO2 is cryogenically separated from water vapor and the bulk air components. From one ice cube, weighing
~30 g, five discrete sub-samples are collected from the continuously released gas stream and
separately stored in small glass tubes. During the ongoing sublimation the ice cube shrinks
nearly symmetrically, which is illustrated in Figure 3-3. In parallel, the corresponding total air
content of these sub-samples is determined manometrically. From this, the CO2 mixing ratio
is calculated using the m/z 44 signal from the mass spectrometric measurement. Off-line coupled to the sublimation extraction is the CF-IRMS system, where the stored tubes with the
CO2 are opened (‘cracked’) within a miniature tube-cracker and the CO2 sample is transferred
in a He carrier (Fig. 3-2). The He stream is first dried and with a cryofocus capillary a sharp
peak is shaped, which is transferred to a GC column. The separated CO2 peak is admitted to
the IRMS via an open split. Both systems are equipped with reference devices to either introduce whole air standards or pure CO2, and thus to mimic the sample’s way through the various analytic steps.
-1
V1
IR lamps
cooling device
LN
V2
ice
sample
V3
PS
sublimation
vessel
LN
flange with V5
copper
gasket
V4
PH
glass
tube
LN
CO2 trap
external water trap
-140°C
reference capillary
-1
0.04 mL min
V12
cooling
jacket and
internal
water trap
V10
V11
V13
V6
V9
V8
LN or +100°C
molesieve
with
resistance
heating
V7
PM
1L
fore
pump
turbo
pump
PVAC
insulated expansion volume
with thermocouple
Figure 3-1 Schematic view of the sublimations extraction apparatus. The sublimation vessel with the ice sample is cooled by cold air supplied from a cooling device. The released air is
dried in the external water trap and CO2 is condensed in the CO2 trap held at liquid nitrogen temperature (within blue dashed line) and transferred into a glass tube for further purification and analysis in the CF-IRMS system. The non-condensable gases like O2 and N2 are collected on the molesieve trap (within green dotted line) to determine the corresponding air
volume. Air standards can be introduced into the sublimation vessel via the reference gas unit (within red dotted line) to provide identical treatment of samples and standards.
large
diameter
water trap
vacuum
pumps
whole air
reference gas
PR1
purge flow 140 mL min
purge flow 0.5 mL min
-1
fill
inject
4.5 cm
Vent
reference
loop 10 µL
vent
6P2
6P1
tube cracker
transfer
clean
mixing
chamber
He out
Nafion
dryer
He
LN
actuator
LN
refilling
device
cryofocus
capillary
GC 70°C
Poraplot
column
25 m
4P
He out
CO2
flowmeter
flow control mode
measurement
He
46
Farraday cups
m/z 44 45
ion source
reference port
open split for std on/off
He in
Nafion
dryer
He
He
Finnigan GP box and IRMS MAT 252
Figure 3-2 Flow scheme of the continuous flow tube cracker-GC-IRMS line. It consists of a CO2 reference gas system (red area) to introduce CO2 pulses onto the tube cracker, followed by
a Nafion dryer, cryofocus capillary, GC, where CO2 is chromatographically separated from N2O and organic impurities. Via the Valco valve ‘4P’ the system is linked to a commercial Finnigan GP box and MAT 252 (yellow area). The GP box houses the open split and the reference port for standard on/off. The open split capillary directs a fraction of the helium carrier to the
ion source of the isotope ratio mass spectrometer (IRMS), where the ion currents of mass 44, 45, and 46 are measured and after integration the δ13C values are computed.
cracker flow
-1
8 mL min He
humidifier
H2O 25oC
GC flow
-1
0.85 mL min
He
He
CO2
Methods and instruments for δ13C and CO2 analysis on ice cores
53
4.3 cm
10.1 cm
diameter of the ice core
3.3 cm
dimensions of the analysed ice sample:
diameter: 3.3 cm
length:
4.3 cm
dimensions of the core segment
reserved for gas analysis:
4.3 x 5.3 x 40 cm
5.3 cm
length of ice sample [cm]
2
1
0
-1
sample
sublimation
increment
duration
[g]
[min]
tube 1:
ca. 3.5
11
tube 2:
ca. 4.5
13
tube 3:
ca. 5.0
15
tube 4:
ca. 5.5
18
tube 5:
ca. 5.0
20
remaining: ca. 6
-2
-1
0
1
radius of ice sample [cm]
Figure 3-3 The top of the figure shows the dimensions of the available ice core segment and the actual size of
the ice cube used for the sublimation. The bottom of the figure illustrates the continuous shrinking of the ice
cube with ongoing sublimation. Note that during the sublimation the ice cube retains its natural orientation, i.e.
annual layers would be horizontal in this figure and orthogonal to the sublimation front. As the outer parts are
gradually removed, the five sub-samples, or tube No. 1-5, originate from increasing ‘depth’ of the ice cube.
Each tube contains on average the gas content of ~5 g ice, which corresponds to ~0.5 mL STP. A small amount
of ice is left over, since the rate of the gas release decreases as the irradiated cross section of the small piece
absorbs less energy.
3.5 The sublimation extraction system
3.5.1
Overall idea behind the sample trapping
The following paragraphs describe the technical details of the sublimation apparatus and its
periphery. First, the procedure of repeated collection cycles is explained. As already mentioned above, one important feature of the presented sublimation apparatus is to process a
single ice sample in a way to yield several sub-samples for the isotopic measurement. To do
this, ice is sublimated continuously and from the liberated gas stream several discrete subsamples are consecutively prepared for the off-line IRMS measurement until the sample is
consumed. The reason for this approach is twofold:
Methods and instruments for δ13C and CO2 analysis on ice cores
54
•
to yield better statistics of sample replicates from the IRMS analysis without opening
the sublimation vessel to the atmosphere to load the next sample. Opening the vacuum system and the subsequent pumping to remove atmospheric CO2 adsorbed on the
surfaces is the most critical step in all applications for CO2 and δ13C analysis.
•
to be able to detect side effects at the beginning and at the end of the sublimation
phase. At the beginning the sublimation system needs some time and sample gas to
equilibrate until stable conditions within the apparatus are reached. The conditions
change again as the rate with which the gas is released from the ice decreases with
sublimation time.
Depending both on the applied strategy for the trapping procedure and also on technical details of the equipment, more or less sub-samples can be obtained from one ice cube. Particularly crucial is, how fast one cycle can be completed to get the system ready for the next cycle. Figure 3-4 describes two collection schemes for the sublimation extraction: an earlier
version, and the final version, which is presented in this work. In both cases trapping of CO2
and air from the gas stream has to be interrupted in order to transfer the collected gases from
the traps (light blue and yellow bars in Fig. 3-4). In the early version of the apparatus, the
sample gas liberated during this time interval was diverted to vacuum until the cold traps were
ready for the next collection cycle. Due to this partial sample loss, only three sub-samples
could be collected from one ice cube. Unfortunately, the δ13C values of the three sub-sample
tubes slightly deviated from each other, which is most likely due to the non-quantitative collection approach (around 30 to 40% of the released gas was directed to vacuum). As isotopic
fractionation between the sublimation vessel and the cold trap under the prevailing vacuum
conditions seemed likely, the system was modified to a achieve quantitative sample recovery.
The final design of the trapping devices allowed to shorten the time interval between two
trapping cycles from six to three minutes. Further, this version improved the volumetric measurement of the air content from which the CO2 mixing ratio is calculated together with the
peak intensity of the IRMS signal. In the final version the sample gas is buffered for a short
time interval (3 min) within the sublimation vessel until the traps are ready for the next cycle.
Due to the accumulation of the liberated gas during the sublimation, the pressure within the
vessel increases until the flow is re-directed to the cold traps (details concerning the precise
timing of this procedure are presented in Chapter 3.7). This allowed a nearly quantitative
sample collection. A small gas loss both at the very beginning of the sublimation and at the
end, when the sample is almost consumed, is inherent to a continuous approach and not
avoidable (red bars in Fig. 3-4). In total, more sample gas is collected this way, therefore five
instead of three tubes can be collected. In the following description of the developed system
only the final version is being presented.
Methods and instruments for δ13C and CO2 analysis on ice cores
55
early approach:
- equilibration step prior each collection cycle, therefore
- sample gas loss to vacuum between sub-samples (around 40%)
- 3 sub-sample tubes
0
10
tube No. 1
30
20
40
tube No. 2
50
60
70
time [min]
tube No. 3
final approach:
- short pressure rise in the sublimation vessel between two collection cycles
- no sample gas loss between sub-samples (quantitative extraction)
- 5 sub-sample tubes
0
10
20
tube No. 1 tube No. 2
30
40
tube No. 3
50
60
tube No. 4
70
80
time [min]
tube No. 5
legend:
collected CO2 transferred in two steps into glass tube for storage
collected air content desorbed, measured, and pumped to vacuum
sample gas directed to vacuum during preconditioning /equilibration
sample gas collected on CO2 trap and molesieve trap
pressure rise in sublimation vessel due to accumulation of released gas
Figure 3-4 Comparison of collection schemes for the liberated air during the sublimation (top: early version;
middle: final approach; bottom: legend). In the early version the entire procedure comprised three collection
cycles, while five cycles in the final version. In both cases, the sublimation of ice samples is a continuous process, releasing a steady gas flow, which is then directed through the two cold traps (blue and green bars), where
CO2 and the air content are trapped separately. At the end of each cycle, the trapping process is interrupted for
gas transfer and to prepare the traps for the next cycle. CO2 is transferred from the trap to the glass tube (light
blue bars), while the air content is desorbed from the molesieve, its pressure measured and the pumped to vacuum (yellow bars). The technical set-up and collection scheme for the early version of the sublimation system
allowed only three sub-samples to be collected, while a large proportion of the sample gas was directed to vacuum between the trapping cycles (red bars). In the final set-up, the sample loss during the interruption for the
gas transfer was solved as the released air during the sublimation is buffered within the sublimation vessel for 3
min (grey bars). This allowed five sub-samples to be collected and prevented sample loss in between.
56
Methods and instruments for δ13C and CO2 analysis on ice cores
3.5.2 Vacuum system and water removal
The vacuum within the sublimation apparatus (Fig. 3-1) is provided by a turbo molecular
pump backed by a rotary vane (both Leybold Vacuum, Germany). Under conditions where no
sample gas is directed to the vacuum line a minimum pressure of typically 2 x 10-7 mbar is
reached at the ‘PVAK’ pressure gauge. A ½’’ o.d. water trap was inserted between the sublimation vessel and the turbo molecular pump to prevent H2O from entering the high vacuum side
via route V2 and V1 (Fig. 3-1). This would be the case, (1) while pumping off the moist atmospheric air after the sublimation vessel was opened to insert the sample, (2) and afterwards,
during a 2-hour cleaning step to remove some mm ice from the outer part of the sample. H2O
strongly adsorbs on the stainless steel tubing and large diameter flexible hoses joining the
vacuum line with the turbo pump, thus, it would take hours to remove all adsorbed H2O from
the surfaces and reach low baseline vacuum levels. Low baseline vacuum levels are crucial to
detect leaks within the sublimation line.
3.5.3 Cooling system for compressed air
As stated above, one crucial feature of the newly developed sublimation line is its compact
design that drastically reduces the surface area of the apparatus. This goal was achieved by
combining the two main processes, sublimation and condensation of the ice, into one single
glass vessel (see Fig. 3-1 and Chapter 3.5.4). To hold the internal water trap, i.e. the upper
part of the glass vessel, at cold temperatures, an air stream from the cooling system is passed
through a cooling jacket (details Fig. 3-1). For this, an insulated PCV tube of 8 cm with an
inner diameter of ~5 mm wider than the glass vessel is used. The small slit between the jacket
and the glass maintains a high and turbulent velocity of the coolant for better heat transfer to
the glass vessel. The cooling jacket is attached and sealed to the flange with an O-ring. The
still cold air stream passing the cooling jacket then cools the lower part of the vessel, which
absorbs energy from the long wave part of the IR lamps.
However, a special cooling mechanism was required to cool the upper part of the glass vessel,
while the lower part is irradiated with IR to sublimate the ice sample shown in Figure 3-1. No
commercially available cooling device was found to suffice these technical demands.
A cooling system was developed that uses compressed air as the coolant medium. The system
was constructed to fulfill the following requirements:
•
to produce ~50 L min-1 cold air at -120 °C, equivalent to ~100 W (assumed ΔT of
coolant and glass vessel ~100 °C)
•
automatic adjustment to the selected temperature set-point of the coolant
Methods and instruments for δ13C and CO2 analysis on ice cores
•
57
to manually adjust the temperature set-point and flow rate during the operation to
meet the changing cooling demand for the sublimation vessel to remove latent heat
flux and cooling of the lower glass section
•
reliable performance during a 3-hour operation period
•
technically adaptable for changed cooling demands during the development of the
sublimation apparatus
Figure 3-5 shows the technical components of the cooling system. Briefly, compressed air is
passed through a coiled heat exchanger unit made of copper tubing, which resides within a
Dewar vessel. With a ‘pump’ a stream of liquid nitrogen droplets (LN) is automatically transferred from a reservoir into the heat exchange Dewar until the given temperature set-point is
reached. Thermocouple 1 (TC1) measures the temperature of the cold air stream that is directed to the cooling jacket of the internal water trap of the sublimation vessel. A PID temperature controller (series 2300, West, UK) compares this temperature with the set-point and
either opens or closes the relay for the electric current of the ‘pump’.
The physical principle behind this simple ‘pump’ is the so called Leidenfrost phenomenon. If
liquid N2 pump
liquid N2
droplet
heat exchanger for compressed air
N2 gas stream
V16
gas
bubble
to cooling jacket
TC1
vent
V15
relay switch
liquid
N2
compressed
air 1 bar
PID
controller
I and II
V14
flowmeter
+ -
manometer
power supply
6 V dc
heating
element
10 W
TC2
liquid N2 storage Dewar
Figure 3-5 Schematic of the cooling apparatus for the sublimation vessel. The system consists of a pumping
device transferring liquid nitrogen (LN) droplets from the 5 L storage Dewar to the heat exchanger unit. Within
the heat exchanger the LN droplets vaporize and cool the copper coil in which the compressed air flows. With
the PID I controller and thermocouple 1 (TC1) the set-point of the produced cold air stream can be adjusted.
For the sublimation of ice samples usually a set-point of -120 °C is used with a volume flow of ~30-50 L min-1
at a pressure of 0.3-0.7 bar. Pressure and flow of the compressed air are manually adjusted using the flowmeter.
58
Methods and instruments for δ13C and CO2 analysis on ice cores
LN gets in contact with a warm surface (ΔT between LN and the object >100 °C) the cold
liquid is quickly rejected from the surface and moves on its own vapor cushion, see Linke et
al. (2006) for details. Hence, the heat transfer from the object to the LN droplet is largely inhibited if the temperature difference remains above a certain limit (Leidenfrost point). The
small heating element at the bottom of the Dewar produces a gentle stream of nitrogen gas,
which transfers the LN droplets through the tubing (see Fig. 3-5). This principle allows to
pump LN droplets at -196 °C through a silicon tubing at near ambient temperature without
vaporizing the LN within the tubing. Within the heat exchanger the LN droplets then vaporize
and produce a large temperature gradient within the Dewar of the heat exchanger. To prevent
overshooting of the cold air device, a second PID controller with TC2 stops the pump when
the threshold is reached. The resulting temperature of the coolant gas measured at TC1 is
fairly constant over time with changes being <2 °C. Although the principle of this simple LN
pump was already used by Brenninkmeijer (1982), yet no application made use of actually
pumping LN through longer tubing to link a LN reservoir with a second external device allowing more versatile applications.
Note that a similar LN pump is also used to transfer LN to the external water trap (Chapter
3.5.6) and to automatically refill the cryofocus trap in the cracker-GC system (Chapter 3.6.6).
To warm up the sublimation vessel after a sample is completed, the cold air stream is interrupted by closing V14 and opening V16 (see Fig. 3-5). Then the compressed air bypasses the
heat exchanger and directly flows to the sublimation vessel to melt the condensed ice. With
this bypass mode, the ice in the sublimation vessel is melted in ~20 min and the water can be
removed to begin the next sample procedure.
3.5.4 Sublimation vessel, internal water trap, and IR lamps
As stated by Güllük et al. (1998) and Siegenthaler (2002), degassing and adsorption of CO2
from glass and metal surfaces and polymer O-rings are critical issues and they reduce the
overall precision of the methods. As pointed out above, the main advantage of our approach
lies in the compact design combining the sublimation of the ice and the close-by removal of
the bulk water vapor into one single glass vessel (shown above in Fig. 3-1). The vessel dimensions are: length 121 mm, outer diameter 33 mm and equipped with a DN40 CF flange to
be mounted to a DN40 double feedthrough (Caburn, UK). On the head of the flange, a pressure transducer (‘PS‘, 100 Torr max, Leybold Vacuum, Germany) is mounted to control the
H2O vapor pressure during the sublimation. Furthermore, an inlet capillary to continuously
introduce a whole air standard is mounted on the feedtrough (see whole air reference inlet in
Chapter 3.5.8).
Methods and instruments for δ13C and CO2 analysis on ice cores
59
The above mentioned dimensions of this glass vessel were chosen for an optimized use of the
available segment of the ice core (see Fig. 3-3). Note that the EPICA cutting scheme for the
gas piece does not allow for larger sample sizes as were available in the older studies, where
large cross sections were used (Fischer et al., 2003).
The total volume of this sublimation vessel with the associated joints and valves amounts to
about 200 mL. For comparison, the devices used by the previous studies were considerably
larger with 1700 mL reported by Siegenthaler (2002). Note that for convenience the volume
of the devices is used here to roughly compare their areas. Since the sample size of Siegenthaler (2002) was comparable to the ~30 g used in this work, with this new approach a considerable reduction of the specific surface area of the apparatus was achieved (i.e. the surface
area of the sublimation apparatus exposed to the sample). Without this considerable reduction
of specific surface area the necessary precision for δ13C and CO2 mixing ratio would not be
feasible.
This single vessel approach results in two benefits: First, the total surface area is reduced and
the number of potentially leaky seals or connections is kept at its minimum. Secondly, it optimizes the flow conditions of the water vapor, i.e. short distances and a large flow crosssection. Consequently, the pressure and temperature differences between the heated ice sample and the condensed ice in the internal water trap is only small due to the high conductivity
of the cross section. This is mandatory to achieve a high sublimation rate at a low temperature
and pressure to keep the ice surface well below -20 °C. Above this temperature a quasi-liquid
layer might form on the ice surface allowing chemical reactions to take place (Güllük et al.,
1998; Barnes et al., 2003).
The glass bulb of the vessel is made of 7052 borosilicate glass that is transparent to visual and
infrared light (λ 0.3 to ~3.5 µm). Infrared light from four halogen bulbs (12 cm length, 500 W
maximum at 230V, Osram) provides the energy for sublimation. The current for the bulbs is
regulated by a continuous power supply to adjust the output to 200-400 W. Compared to the
IR laser source used by Monnin (2000) and Siegenthaler (2002), here the simpler approach
with halogen bulbs already used by Güllük et al. (1998) and Wilson and Long (1997) was preferred, as this set-up avoids problems associated with the optical window, e.g. leak problems
due to temperatures change and only a partial sample consumption due to the unidirectional
irradiation by the laser.
3.5.5 External water trap
Although the internal water trap already removes 99% of the water vapor, a special external
water trap was needed to achieve the requirements for extreme low pH2O. Furthermore, a
compact design to minimize the total surface area of the system was also aimed at. In most
60
Methods and instruments for δ13C and CO2 analysis on ice cores
dual-inlet and continuous flow applications (Werner et al., 2001; Ribas-Carbo et al., 2002;
Leuenberger et al., 2003) the sample gas stream is sufficiently dried by immersing a capillary
coil into a dry ice/ethanol bath of about -80 °C. Since in this work the trapped CO2 and simultaneously traces of H2O are ultimately transferred into a miniature glass tube with a volume of
only 15 µL, an H2O amount of just 0.1 µg is sufficient to form a liquid phase. A rough calculation assuming laminar flow reveals that this amount is already co-transported to the CO2
trap at a trap temperature of -100 °C. Note that the released air amount of ~0.5 mL STP per
sub-sample occupies a volume of ~5000 mL at the operating pressure of ~0.1 mbar within the
traps. The actual H2O flux is further increased if a diffusive flux by a Knudsen flow regime
(mean free path at 0.2 mbar ~1 mm; CO2 trap i.d. 2 mm) is assumed instead of a laminar flow.
In cases the H2O flux exceeded 0.1 µg, liquid H2O formed within the tube after the transfer
from the CO2 trap to the glass capillary and warming to ambient temperature. Further, the
presence of a liquid phase within the tube leads to an isotopic exchange of oxygen between
CO2 and H2O. This shifts the δ18O values to more depleted values of up to 10‰. Therefore,
temperatures as cold as -140 °C in the water trap are needed to suppress the formation of a
liquid phase in the tube. Since classical cooling systems, e.g. dry ice/pentane slush (Leckrone
and Hayes, 1997) are not readily suitable for this temperature range and closed-cycle He
coolers are too bulky, a special trap cooled with LN droplets and cold nitrogen gas was developed. A second reason for the usage of a LN cooled water trap is the exclusion of solvents
within the laboratory as traces of organic contaminants interfere with the IRMS measurement
(Francey et al., 1999; Leuenberger et al., 2003; Eyer, 2004).
Function and operation of the external water trap:
The technical details of the external water trap are illustrated in Figure 3-6. Similar to the
cooling system for the compressed air, droplets of LN are automatically pumped from a large
reservoir into the trap, where the LN droplets vaporize on an extensive sheet of copper wire.
In this application the silicon tubing that connects the LN pump with the trap has to be heated
with a heating jacket. This was necessary since here LN is pumped upwards (~30 cm) from
the reservoir to the above water trap. As the heat transfer from the tubing to the LN droplets is
more intense in this case, the temperature of the tubing falls below the threshold of the Leidenfrost point and the pump stops operating properly. The heating jacket around the silicon
tubing prevents this and allows upward pumping of LN to the external water trap (the heating
jacket consists of ~1 m resistance wire (10 Ω m-1) wrapped around the silicon tubing, marked
red in Fig. 3-6).
The temperature of the trap (¼’’ o.d., 0.53 cm i.d., 20 cm length) is automatically controlled
via a thermocouple plugged into a PID controller, which regulates the operation of the pump
similar to the air cooling system of the sublimation vessel. On the downstream side of the
Methods and instruments for δ13C and CO2 analysis on ice cores
61
trap, a pressure transducer (PH, 10-4 -1 Torr max) is installed to monitor the gas flow during
the sublimation. Since the trap is operated only 20 °C above the CO2 saturation pressure during sublimation conditions, cold spots in the trap are of special concern. To detect potential
cold spots, the trap was filled with CO2 at 0.001 mbar and the trap was held at -140 °C for 10
min to observe a potential pressure drop due to condensation or adsorption of CO2 onto the
surfaces. Within the range of precision of the pressure measurement no loss of CO2 was observed, thus, there is no indication that the trap shows cold spots.
Note that an exchangeable joint in the silicon tubing of the pump allows alternatively to direct
warm pressurized air through the external water trap (Fig. 3-6). This is necessary to rapidly
warm up the water trap to remove the condensed ice within the trap after the sample is finished. As the external water trap is highly insulated with several layers of insulating foil heat
conduction from the outside would be too low to warm up the trap in a reasonable time.
external water trap
LN pump with heating jacket
pressure
gauge
PH
moist gas tream from
sublimation vessel
flexible joint
for rapid exchange
1 0 V dc
V4 dried air stream
V3
to CO2 trap
thermocouple
stainless tube ¼´´ o.d.
outer insulating layer
copper wire
liquid
N2
deposited ice
aluminium foil
PID
controller
interruptor
heating
element
10 W
inner insulating layer
silicon tubing perforated for
nitrogen release
+ power supply
6 V dc
LN storage Dewar
Figure 3-6 Schematic view of the external water trap with its associated pumping system and temperature
controller. Liquid nitrogen is pumped from the storage Dewar to the watertrap using a heated silicon tubing.
Within the external trap the LN droplets vaporize on layers of copper wire wrapped around the ¼’’ tubing from
the sublimation vessel. The large copper layers conduct the temperature to the inner part of the trap, where the
temperature fluctuations become attenuated. Due to the large heat capacity of the copper layers the trap needs
about one hour to establish isothermal conditions of -140 °C in the inner part (monitored by the thermocouple).
The copper layers are covered with layers of aluminum foil to prevent condensation of atmospheric moisture
that would quickly clog the trap. Several layers of insulating foil wrapped around the trap reduce heat conductance from outside. The flexible joint within the silicon tubing connecting the pump with the water trap enables
a rapid warm up of the trap by flushing it with pressurized air at ambient temperature to remove the condensed
ice after the operation.
62
Methods and instruments for δ13C and CO2 analysis on ice cores
3.5.6 CO2 trap and glass capillaries for CO2 storage
The dried gas stream (H2O dewpoint -140 °C) which passed through the external water trap is
directed to the U-shaped stainless steel tubing (⅛’’ o.d., 0.085’’ i.d., 10 cm length) immersed
in LN (Fig. 3-7). Within this trap, CO2 together with N2O are efficiently condensed from the
gas stream, which is then directed to the molesieve trap to remove the remaining gas components.
To achieve efficient trapping on the CO2 trap, a ⅛’’ tubing was used here instead of ¼’’ tubing used for all other sections within the sublimation and trap system. Without this short ⅛’’
section, the pressure of the gas flow within the CO2 trap would be too low to condense CO2
efficiently. The reason for this is that with decreasing total gas pressure the mean free path of
molecules increases, i.e. molecules travel larger distances between collisions with other molecules or the cold surface of the trap. Further, also the partial pressure of CO2 would then leave
the region where CO2 can be easily condensed at liquid nitrogen temperature. On the other
hand, a too long ⅛’’ tubing section would cause an undesirably large back pressure and long
residence time of the released gas within the sublimation vessel and the external water trap.
This situation has to be avoided for two reasons: First, loss processes, like adsorption of CO2
on surfaces and co-deposition of CO2 with H2O in the water trap, increase with higher CO2
partial pressure. Secondly, the amount of gas remaining in the sublimation vessel at equilibair - CO2 , N2O
1/4'' union Tee and 1/4'' tubing
V6
dryed air stream
from H2O trap
to molsieve trap
V4
heating jacket
1/4'' - 1/8'' reducing union
V5
1/8'' tubing
1/8'' Ultra torr adapter
with
Viton O - ring
LN
glass capillary
1/4'' VCR adapter with copper gasket
LN
Fujikin all metal valve
Figure 3-7 Scheme of the CO2 trapping device. Shown are the positions of the Dewars for the situation during
the trapping of released air with the CO2 trap immersed in LN, while the glass capillary is not immersed.
Methods and instruments for δ13C and CO2 analysis on ice cores
63
rium flow conditions scales with back pressure thereby counteracting with the goal to achieve
quantitative sample collection. Hence, length and shape of the ⅛’’ tubing of the CO2 trap was
optimized to be as short as possible, while at the same time reach high efficiency and a suitable shape to be immersed in a LN Dewar. As the viscosity of gases roughly increases with
the square root of temperature, the sections of the ⅛’’ tubing that are not immersed in LN
during trapping have to be minimized. The optimal solution was found using a U-shaped trap
of ⅛’’ tubing with 10 cm length, which can be immersed in LN just a few mm below the ⅛¼’’ reducing union fittings (Swagelok, stainless steel). With this design almost ~9 cm of the
total length of 10 cm are actually immersed in LN to condense CO2 off the gas stream while
only ~1 cm of the ⅛’’ tubing are above the LN level. A heating jacket keeps the two ⅛-¼’’
reducing unions at ambient temperature while the ⅛’’ tubing of the trap is immersed in LN
(Fig. 3-7). If not heated, the unions would cool down to almost LN temperature while the trap
is immersed in LN and due to their larger thermal inertia compared to the thin ⅛’’ tubing,
warming-up would be considerably retarded during the CO2 transfer into the glass tube.
Attached to this trap is a long glass capillary, to which the trapped CO2 is transferred and then
a section of 2 cm of this capillary is flame-sealed and the closed tube is stored for the later
steps (see Chapter 3.6). Hence, the glass capillary and the resulting tubes act as the link in this
off-line coupling.
As tube material standard Pasteur pipettes are used (length of the capillary 12 cm, o.d. 2 mm).
Prior to usage the glass capillaries are cleaned in an ultrasonic bath with diluted HCl and
thoroughly rinsed with deionized water to eliminate any organic and inorganic contaminants
from the glass surface. The capillary’s tip is flame-sealed and the outer diameter of the open
end is adjusted to 3 mm and rounded with a hand torch to fit a ⅛’’ o.d. Cajon-Ultratorr
adapter. With this adapter the glass capillary is connected to the ¼’’ tubing leading of the V5
valve, which links the CO2 trap with the glass capillary (see Fig. 3-7).
3.5.7 Molesieve trap - air content measurement
Although the main purpose of the described analysis method of this work is the precise determination of δ13C, measuring the CO2 mixing ratio on the same sample is very useful. First,
it is a valuable tool to detect contamination or loss processes during the whole analysis. Secondly, for a quantitative interpretation of global atmospheric δ13C changes, it is imperative to
have the data of both the isotopic and the mixing ratio at the same time interval. Although
highly resolved time series on CO2 mixing ratio are available for the Holocene (Monnin et al.,
2001; Siegenthaler et al., 2005a), temporal resolution is still poorer during MIS3 and older
periods. Further, uncertainties with dating or a mismatch in the ice-age/gas-age difference of
two cores deteriorate the precision necessary to entangle global carbon fluxes.
64
Methods and instruments for δ13C and CO2 analysis on ice cores
With the mass spectrometer providing the amount of CO2 via the m/z 44 peak intensity, the
corresponding amount of air is determined volumetrically with the molesieve trap. Briefly, a
glass tube (¼’’o.d, ⅛’’ i.d.) is filled with 5 Å molesieve pellets. Before each use and for
weekly regeneration, water is removed by heating the molesieve to ~200 °C for some hours.
During the sublimation, the molesieve is immersed into LN and acts like a normal vacuum
pump. At -196 °C the equilibrium pressure of a loaded molesieve (1 mL STP air) is <0.0020
mbar (PM pressure transducer, 10-4 - 1 Torr max). It is therefore the molesieve ‘pump’ which
drives the pressure gradient from the sublimation vessel through the water trap and the CO2
trap to the molesieve trap. Typical values during the sublimation of an ice sample are ~0.4
mbar within the sublimation vessel, whereby the partial pressure of H2O dominates while the
partial pressure of the air constituents account for only ~0.02 mbar. However, after the external water trap, pressure reading at the PH transducer, the total pressure of ~0.09 mbar is entirely established from the air constituents as H2O is removed from the gas stream. A pressure
of only ~0.02 mbar is measured in the molesieve trap with the PM transducer.
The quantitative release of the adsorbed air is accomplished by quickly heating the molesieve
to 100 °C using an automatically controlled heating jacket. This device allows heating the
molesieve from -196 °C to +100 °C within 40 s using a resistance wire (length 50 cm and 10
Ω m-1 operated with 7 V). After the heating jacket is switched on, the temperature is rapidly
rising until the set-point is reached at the thermocouple and the PID controller interrupts the
current and adjusts the temperature at +100 °C; i.e. a technical set-up similar to the external
water trap. Note that the quick desorption of the air from the molesieve trap is of special importance since this step determines the time the trapping is interrupted and the released air
from the sublimation buffered within the vessel. The desorbed air from the molesieve expands
into an insulated 1L stainless steel cylinder (referred further on as expansion volume) and the
temperature of the expansion volume is read out using a thermocouple (see Fig. 3-1). Insulation of the expansion volume and monitoring its actual temperature is essential since temperature fluctuations would otherwise bias the pressure at the PM transducer, thus, the calculated
CO2 concentration. After the desorption from the molesieve, the air occupies the entire volume between the valves V6 and V9, framed by the green rectangle in Figure 3-1. However,
only the temperature of the insulated expansion volume is sufficiently stable and is measured
for the temperature correction of the pressure. Especially the temperature of the tubing section
between V6 and V7 cannot be stabilized fast enough. However, the enclosed volume of this
section is only ~20 mL, thus, small compared to the 1000 mL of the temperature controlled
expansion volume and negligible within the measurement precision of temperature and pressure.
A typical ice sample (~6 g per sub-sample) with a gas content of about 0.5 mL STP results in
a pressure of 0.5 mbar within the expansion volume. After the molsieve trap reached the set-
Methods and instruments for δ13C and CO2 analysis on ice cores
65
point of +100 °C at ~40 s, the desorption is quantitative and the pressure becomes stable
(trend <0.0005 mbar min-1) at the time of read out at 60 s. This design allows a precision of
the total air content measurement of <0.5%, which corresponds to <1 ppmv for the CO2 mixing ratio.
3.5.8 Whole air reference inlet to introduce whole air standards
The prerequisite of accurate measurements for isotope analysis is the so called ‘principle of
identical treatment’, coined by Werner and Brand (2001). To fulfill this requirement, one
would need at best artificial ice with ~10% air inclusions of known composition.
This reference material is not available as this would imply to reproduce the sintering conditions of snow to bubbly glacier ice in the laboratory. At low temperatures this close-off process takes at least decades to proceed (see Chapter 2.1). Speeding up this process about several
orders of magnitude would either imply increasing the temperature near to melting point and
working with a high pressure apparatus. As an isotopic fractionation of the enclosing air is
very likely to occur under these conditions, a thus produced reference material would not be
trustworthy enough for the aimed high precision measurements.
The next best referencing strategy is a whole air standard introduced during the sublimation of
a blank ice cube. This treatment mimics the air release from the sample during the sublimation as closely as possible. Except for the actual gas release off the ice surface, all subsequent
steps are identical for reference and sample.
For the calibration of the system a high pressure cylinder with synthetic air (Air Liquid, Germany) is used as the whole air working standard (referred as ‘air standard’). Its precise gas
concentrations and the isotopic composition of CO2 have been measured at the Institute for
Environmental Physics at the University of Heidelberg by I. Levin (CO2 = 277.7 ppmv; δ13C
= -2.75‰; δ18O = -14.72‰; N2O = 233 ppbv).
Since the release of the trapped air during sublimation is a slow and continuous process, a
reference inlet was needed to introduce a constant flow of compressed air into vacuum conditions without fractionation. Admitting air into vacuum without isotopic fractionation is not
trivial. Generally, during the expansion from the high pressure region to vacuum conditions
the physical principles of gas flow change dramatically. With decreasing pressure three flow
regimes are distinguishable: (1) viscous flow, (2) Knudsen flow, and (3) molecular flow. The
prevailing regime at a given point can be classified by the ratio of the molecules’ mean free
path to the cross section diameter of the tubing. Pure viscous flow occurs below a ratio of 0.1,
and pure molecular flow above a ratio of 10, with Knudsen flow in between. Both Knudsen
flow and molecular flow cause large a fractionation (up to 4.4‰ for diffusion of CO2 in air)
66
Methods and instruments for δ13C and CO2 analysis on ice cores
of the gas species and their isotopes since diffusion of the molecules is mass dependent. This
produces elemental and isotopic gradients within the tubing, which was already noticed in the
early days of mass spectrometry (Honig, 1945; Halsted and Nier, 1950).
Consequently, the device to transfer gas from a high pressure reservoir to a vacuum apparatus
must avoid net fractionation between the reservoir and the admitted gas. This can be achieved
when the flow velocity within the tube is high enough to overcome any back diffusion to the
reservoir. The developed reference gas inlet system consists mainly of four components (see
section enclosed by red rectangle in Figure 3.1):
(1) reservoir with the compressed air standard
(2) pressure regulator
(3) inlet capillary
(4) purge port with two vent capillaries
(1) The working standard is stored in a 1000 mL stainless steel container at 500 kPa and is
filled from the high pressure cylinder. (2) With a pressure regulator (PR1; Porter 8286SMVS-30, USA), the air pressure and thus the flow through the capillary can be adjusted
from 100 to 350 kPa to cover the rates of air release during the sublimation of ice samples
(0.02-0.06 mL STP min-1). (3) The inlet capillary has an i.d. of 0.05 mm, thus, the same i.d. as
the CO2 reference gas capillary of the MAT 252 and provides a viscous flow regime at this
pressure range. The length of the capillary is adjusted to 2.5 m to attain flow rates of 0.020.08 mL STP min-1 at corresponding pressures of 100 to 350 kPa. Usually, an air pressure of
250 kPa is set at the pressure regulator (PR1) leading to a flow rate of 0.04 mL STP min-1.
This rate is comparable to the rate with which air is released during the sublimation of ice
core samples. The exit of this capillary within the sublimation vessel is only a few mm above
the ice cube to release the air standard in a position similar to the sample. (4) The volumetric
flow through this inlet capillary is too low to flush the reference system efficiently. Note the
internal volume of the entire reference device amounts to ~ 7 mL, which is largely due to the
dead volume within the pressure regulator. Therefore, the reference system is equipped with a
purge port continuously directing air to the vent capillaries. Via a 4-port valve (V13,Valco,
SS2H) the air standard can be either directed to a wide bore capillary (fused silica, 0.23 mm
i.d., 1 m) for a rapid high volume flushing allowing for approx. 140 mL STP min-1 at 200 kPa.
Or alternatively, to a narrow bore capillary (fused silica, 0.05 mm i.d., 15 cm) with which the
system can be continuously flushed during preparation of air standard tubes. This flow is ~0.5
mL STP min-1 at 200 kPa. Permanently flushing the inlet system is a prerequisite for stable
results, otherwise long lasting drift phenomena for 13C and CO2 and bad precision would result.
Methods and instruments for δ13C and CO2 analysis on ice cores
67
The inlet system is connected to the vacuum system via V10. This valve is kept closed when
the reference inlet is operating and air standard being introduced to the sublimation vessel via
the inlet capillary. Contrary, when an ice core sample is sublimated within the vessel, V10 is
open to the vacuum side to evacuate the inlet system, with V11 and V12 being closed. The
outer diameter of the union cross of the inlet system and the tubing between the ⅛’’ valves
V10, V11, and V12 (Swagelok SS-2H) is ⅛’’ and connections are kept as short as possible,
typically 3 cm. This helps to minimize the internal volume of the inlet and reduces the residence time of the gas within the reference system.
Considerations for the preparation of air standards:
With this reference inlet an air standard can be introduced during the sublimation of bubble
free blank ice, thus, identical treatment of sample and reference is possible. However, at best
two blank ice samples a day can be processed this way, resulting in ten single tubes. This is
due to the fact that the sublimation of one blank ice sample results in five air standard tubes
and a cycle of opening/closing the sublimation vessel followed by long pumping takes about 3
hours.
To overcome this inherent drawback, additionally a more rapid procedure is used to provide
air standard tubes. This was useful to test the performance of the system at different temperature, pressure, or flow conditions during the development of the apparatus. Instead of admitting air while the blank ice is actually sublimating using the IR lamps, ice is held at the same
temperature and pH2O conditions as during the actual sublimation, i.e. -27 °C corresponding
to a H2O partial pressure of ~0.5 mbar. However, the blank ice is not ‘consumed’ since the
sublimation vessel is held isothermal without temperature differences between blank ice and
cold air steam. Hence, using this procedure termed ‘static’, the time consuming replacement
of the ice is not necessary. With this procedure, pressure and flow within the traps are identical to the conditions during sublimation, yet the H2O pressure gradient and the H2O flux
within the sublimation vessel is lacking. Further, the glass vessel is not irradiated during this
procedure (see Chapter 3.7.2 for a detailed description of air standards). Both approaches
were used during the development of the analysis as well along with the processing of ice
core samples (see Chapter 3.9 for the results of air standard measurements).
3.6 The tube cracker-GC-IRMS system (CF-IRMS)
The following paragraphs deal with the development and functionality of the tube cracker-GC
system (Fig. 3-2). First, the reasoning for a gaschromatographic purification of the extracted
gas sample prior to the IRMS analysis is stressed, followed by the general idea of the meas-
68
Methods and instruments for δ13C and CO2 analysis on ice cores
urement scheme. Finally, the technical components necessary to realize this scheme are described.
Briefly, the tubes with the extracted gas sample, consisting mainly of CO2 and N2O, are
opened (‘cracked’) using a tube-cracker device and the sample is entrained in a He carrier.
The gas stream is dried, cryofocused, and CO2 is separated from N2O and other impurities
using gas chromatography. The purified CO2 peak is then introduced to the ion source of the
IRMS, where the ion currents of mass 44, 45, and 46 are measured.
3.6.1 Reasoning for a GC separation of the extracted gas sample
As outlined above, the entire analysis method for measuring δ13C on ice core samples was
split into two separate lines, thus, off-line coupled. Within the vacuum line of the sublimation
apparatus, CO2 and N2O are separated from the bulk air components (N2/O2/Ar) using cryogenic traps. Afterwards, gas chromatography is used to separate CO2 from N2O and organic
contaminants.
N2O is isobaric to CO2, i.e. within the ion source of the IRMS it produces ions with masses
44, 45 and 46 like CO2 does, but at different isotope ratios. A typical mixing ratio of CO2/N2O
in air of ~1000 causes a depleting of the δ13C of CO2 by about 0.3‰. For dual-inlet measurements this effect can be corrected mathematically when the ionization efficiency of CO2 vs.
N2O is empirically determined and the CO2/N2O mixing ratio is well know as it is the case for
modern atmospheric air samples (Ghosh and Brand, 2004). For ice core samples, however,
this correction is not easily possible. First, due to the sparse N2O data at glacial times, and
secondly, due N2O in-situ production in dust-rich glacial ice samples (Spahni et al., 2005).
While GC separation of CO2 and N2O was already used for measuring δ13C on atmospheric
air samples (Ferretti et al., 2000; Ribas-Carbo et al., 2002), for ice core samples this technique is applied here for the first time (see Chapter 3.6.5).
As discussed in Chapter 3.2, nearly all authors who analyzed δ13C on deep ice cores reported
problems during the mass spectrometric analysis caused by drill fluid contamination 19. Since
some components of the drill fluid behave physicochemically similar to CO2 during the extraction process, minute traces can simultaneously reach the ion source together with CO2.
During the course of the early development for this work, ice core samples were measured
without GC separation resulting in excessively high m/z 45 and 46 traces yielding highly enriched δ13C ratios. As an example, Figure 3-8 shows the measurement of a contaminated ice
core sample. A δ13C value of +1420‰ was calculated for this sample illustrating the high
sensitivity of even small amounts of contaminants in ice core samples. Recently, Eyer (2004)
demonstrated that probably the halogenated compound C2H3Cl2F, or Freon141b, used as den19
note, the use of drill fluid is compulsory during the operation of deep ice core drilling as otherwise the open
drill hole would rapidly close due to ice flow. Contrary, shallow cores can be drilled ‘dry’.
Methods and instruments for δ13C and CO2 analysis on ice cores
69
sifier in the drill liquid is producing these m/z 45 and 46 fragments during its disintegration in
the ion source. Although a severe contamination can be easily detected by unusual positive
δ13C values and excluded from the data set, minor contaminations are harder to detect and
require many replicates or a comparison of different cores. As a conclusion, gas chromatographic separation of CO2 from N2O and organic impurities from the extracted gas sample
is indispensable.
pure CO2
ref erence
peak
m/z 46
m/z 45
m/z 44
2000
2
ratio 45/44
1000
ratio 45/44
Intensity [mV]
3000
4
ice core sample
contaminated w ith drill
fluid
3
1
1500
1550
1600
1650
Time [s]
1700
1750
0
Figure 3-8 Example of a contaminated ice core sample (right) with a reference peak of pure CO2 (left). The
heights of the three mass traces of 44, 45 and 46 of the pure CO2 peak reflect the natural abundance of CO2
level with a ratio of 45/44 around 1.1 (right axis). In contrast, the contaminated sample shows highly elevated
intensities of the mass traces 45 and 46 relative to 44, which produce the odd 45/44 ratio on the right axis. Note
that the plotted voltage intensities for the mass traces 44, 45, 46 do not reflect the true abundance of these ion
fragments. For convenience, the measured ion currents were converted to voltages using different resistors to
produce signal ratios near unity (see Chapter 3.6.7).
3.6.2 General layout of the measurement sequence for the CF-IRMS system
The central principle of any stable isotope analysis is to compare the signal of a sample with
that of a reference standard and to report the relative difference to this standard. The reason
for this is that isotope ratio mass spectrometers are especially designed to compare signal ratios at an utmost precision, but not to precisely measure their absolute intensities (Brand,
1996). The key is to treat sample and standard in the course of the analysis in exactly the same
way. This principle of identical treatment is applied during the classic dual-inlet measurement, commonly used for larger sample sizes. In an alternating sequence, the introduction of
samples and standards to the ion source is rapidly switched many times, whereby their signal
intensities are adjusted. While this rapid switching minimizes the temporal drift of the mass
70
Methods and instruments for δ13C and CO2 analysis on ice cores
spectrometer, the adjustment of the signal intensities minimizes the dependence of the δ13C
value on the signal intensity (‘linearity’).
In contrast, accuracy and precision of CF applications are often diminished by the drift and a
‘linearity’ of the mass spectrometer or the periphery since the dual-inlet principle can only be
roughly mimicked. The reason is that in CF applications short sample peaks arrive with the
carrier gas flow at the ion source of the IRMS, instead of a continuous plateau in case of the
dual-inlet. An active adjustment of signal intensities of standard and sample is not possible.
To mimic the dual-inlet measurement principle for the CF approach developed in this work
the following requisites have to be met:
(1) introduce reference peaks onto the cracker and treat them identically to a sample
(2) generate reference peaks with different intensities to cover the range of samples
to correct for linearity
(3) make a single sequence of standard and sample as short as possible to minimize
drift and allow measuring a large number of samples in a single session to profit
from identical measurement conditions
(4) allow standards to be measured in a fully automated procedure to improve reproducibility with statistics and reduce personal attendance to the system
Based on these requirements, the tube cracker GC-IRMS system (shown in Fig. 3-2) was built
and a measurement protocol for alternating standards and samples was developed (Fig. 3-9).
To meet requirement (1) and (2), three different sizes of CO2 pulses can be injected to the
tube cracker via the reference loop (‘linearity loops’ L1, L2, and L3 in Fig. 3-9) to correct for
the sample size effect (see Chapter 3.8). With the He carrier the CO2 sample is then transferred to the cryofocus and to the GC, from which a purified CO2 peak is directed to the ion
source of the IRMS, where the δ13CO2 value is measured. In between, the cracker is cleaned
and prepared for the next sample or reference standard. A pair of one reference peak (EQ, see
Fig. 3-9) and one sample (SA) takes less than 10 min to be completed allowing the repeated
measurement of a large number of reference and sample pairs within a measurement session,
requirement (3). Except for one step, the cracking of the sample tube, the entire analysis with
the tube cracker-GC-IRMS system has been fully automated using pneumatic valves and actuators operated with a script embedded into the IRMS software, requirement (4).
The aim of preceding each linearity or sample peak with an equilibration peak (EQ) is to provide identical measurement conditions by equilibrating both the cracker-GC system and the
ion source. Equilibrating the system always to the same CO2 level is especially mandatory to
achieve a reproducible signal intensity of mass 44 used to calculate the CO2 mixing ratios.
The peak area of a sample peak would otherwise be slightly influenced by the intensity of the
Methods and instruments for δ13C and CO2 analysis on ice cores
71
preceding peak, i.e. affected by an memory effect. Further, the time series of all EQ peaks are
used to calculate the δ13C drift with time (see Chapter 3.8 for data processing).
As can be seen in Figure 3-9, in addition to reference and sample peaks, another type of CO2
peak is admitted to the ion source at the very beginning and at the end of the measurement
run. These std on/off pulses are fundamentally different to the other peaks as they are directly
introduced to the ion source of the IRMS with a separate reference port and a second inlet
capillary (see yellow box with Finnigan GP box and IRMS in Fig. 3-2). These std on/off
pulses do not reflect the conditions within tube cracker-GC system, but only those of the pure
IRMS measurement. Consequently, its principal use is to check the performance of the IRMS
and together with the reference peaks admitted to the tube cracker system they allow to differentiate which part actually causes an observed δ13C drift (see Chapter 3.8).
Figure 3-9 Measurement scheme for the tube cracker-GC-IRMS system. With std on/off reference pulses the performance of the sole IRMS machine without the tube cracker periphery is monitored before and after the actual
tube cracker measurements. The performance (drift, linearity and reproducibility) of the tube cracker system itself
is checked before (blue section) and after (green section) the sample tube measurements in a repeating sequence of
reference peaks with different peak heights (linearity loops L3, L2, L1) to allow correction of linearity effects.
Bracketed between these two linearity sections is the sample section, where the glass tubes with the extracted CO2
from the ice core samples are processed. Each sample tube (SA) is preceded by a reference gas peak (EQ) to
equilibrate the system identically for all tubes measured within a measurement session. Note that the SA tubes
include both ice core samples (SAice) and air standards (SAair) and tubes were randomized prior to the measurement
to cancel out any bias during the measurement run. Typically around 20-30 tubes were measured together resulting
in a total length of the IRMS run of up to 10 hours.
Methods and instruments for δ13C and CO2 analysis on ice cores
72
3.6.3 The tube cracker
The tube cracker device is the link between the sublimation extraction at vacuum and the CFsystem at pressurized He. Within the tube cracker, the glass tube with the extracted gas is
opened by bending it and its content is entrained into a continuous He carrier (Fig. 3-10). Until now, tube crackers were only applied in conjunction with dual-inlet or micro volume IRMS
applications. The traditional design of tube crackers is a dead-end construction of two UltraTorr fittings with a flexible, corrugated steel tubing in between (Des Marais and Hayes, 1976;
Coleman, 1981; Norton, 2005). Tube outer diameter dimensions normally used were ¼’’ or
⅜’’, and after breaking the tube at vacuum the gas was expanded into a bellow or condensed
into a cold finger (Kennedy and Kennedy, 1994). As the latter techniques require rather large
sample sizes (~10 to 100 times more than for CF approaches), use Ultra-Torr fittings with
Viton O-rings, and operate at vacuum conditions, the traditional tube cracker design had to be
adapted to meet the needs of this CF application:
•
a flow through design and a small internal volume to permit rapid and quantitative
transfer at a low flow rate
•
a all-metal stainless steel construction and exclusion of polymer materials to minimize memory and surface effects
The tube cracker developed in this work is composed of a Swagelock reducing union, a 4.5
scored glass tube with CO2 sample
1/8'' o.d. stainless steel tubing
1/8''-1/16'' o.d. column end fitting
1/16''-1/8'' o.d. reducing union
silver wool
stainless steel frit 2 µm
He flow from
via 6P2 valve to
6P2 valve
cryofocus and GC
glass tube is cracked by
bending the cracker
opened glass tube with
fine particles
Figure 3-10 Schematic of the tube cracker device constructed to open small glass tubes in a continuous Helium
flow. The top shows the cracker with an intact glass tube containing the CO2 sample (yellow). Below, the
cracker is shown after bending it a few degrees. The glass tube is opened and the released CO2 being flushed
out of the cracker. For comparison, the length of the glass tube is ~25 mm with an outer diameter of ~1 mm.
Methods and instruments for δ13C and CO2 analysis on ice cores
73
cm long stainless steel tubing, which is the flexible part and houses the glass tube, and a
Valco column end fitting (Fig. 3-10 for technical details). To prevent glass particles from entering the down stream valves, the Valco column end fitting is equipped with a 2-µm stainless
steel frit. A clew of silver wool fixes the glass tube within the cracker and facilitates the
cracking of the tube. The tube is cracked while the cracker is manually bended using two pliers. The total internal volume of the cracker is only 160 µL, thus, only a single supply of He
at a low flow rate (0.85 mL min-1) is required to flush the sample to the cryofocus capillary.
With two stainless steel capillaries the cracker is connected to the Valco 6P2 valve to allow
for flexible handling during the cracking.
3.6.4 Device to introduce CO2 reference gas to the tube cracker
To allow the identical treatment of samples and reference gas standards within the crackerGC-IRMS system, pulses of CO2 in He can be injected into the cracker. The reference gas
assembly consists of three parts (red area in Fig. 3-2):
•
mixing device to dilute CO2 with He
•
stainless steel reference loop (10 µL)
•
Valco 6-port valve to either fill or inject the reference loop
With two pressure regulators (Porter 8286-SMVS-30, USA) the flow rate is set to ~0.1 mL
min-1 for CO2, and ~15 mL min-1 for He. Prior to their use, the two gases are homogenously
mixed within a mixing chamber (15 mL stainless steel cylinder). By changing the flow rates
of the two gases, the CO2 concentration of the He/CO2 mixture can be adjusted to a convenient value of ~1% CO2, or ~2 nmol per 10 µL. This mixing device allows an adjustment of the
signal height of the reference peaks without the need to switch between different loop sizes.
To introduce a CO2 pulse to the tube cracker, the Valco 6P1 valve is switched from the ‘fill
position’ to the ‘inject position’ and the GC flow (0.85 mL min-1) flushes the CO2 from the
reference loop via the 6P2 valve (in ‘transfer mode’) to the cracker device (Fig.3-2). The CO2
reference gas used in this device originates from the same high pressure cylinder as for the
reference port to introduce std on/off CO2 pulses (see Chapter 3.6.7). Its δ13C value is -6.97‰
vs. VPDB, thus close to the composition of the preindustrial atmospheric CO2 value. Note
that during the IRMS measurement and during the data processing the raw δ13C values of the
tube samples are referenced onto this CO2 standard. However, during the final calculation the
δ13C values of the ice core samples were referenced onto the whole air standard (see Chapter
3.8).
74
Methods and instruments for δ13C and CO2 analysis on ice cores
3.6.5 Humidifier for the He carrier (GC flow)
High precision δ13C measurements on CO2 require low and constant water levels within the
ion source of the mass spectrometer. During the ionization, H2O and CO2 form HCO2+ ions
that are isobaric to
13
CO2 and cause an apparent sample enrichment (Leckrone and Hayes,
1998; Meier-Augenstein, 1999; Rice et al., 2001). Therefore, the water vapor level of He carrier in CF-IRMS applications are generally kept as low as possible. Contrary to this common
notion, a special humidifier device that saturates the carrier gas with H2O had to be inserted
upstream to the tube cracker. Note that the water vapor is removed from the carrier via a
Nafion dryer after the carrier passed the cracker (see Fig. 3-2). During an early development
stage of this method, CO2 pulses admitted to the cracker via the reference loop resulted in
reproducible δ13C values, unfortunately, similar measurements of CO2 prepared in glass tubes
revealed a serious fractionation with poor precision (>1‰). The cause of this problem was
identified as a strong isotopic fractionation occurring on the fresh glass surfaces after breaking
the tubes in the cracker.
To do this, empty tubes were prepared and sealed off at high vacuum. These tubes were then
inserted into the cracker and a first measurement was conducted, whereby a first CO2 pulse
from the reference loop passed the cracker flowing around the intact glass tube. Then, a second CO2 pulse was admitted to the cracker while the empty tube was broken some seconds
prior to the arrival of the CO2 pulse at the crushed glass particles. Whereas no effect was visible in case of the intact tube, two effects were visible for the crushed tube. First, the area of
the CO2 peak was considerably reduced by up to 15%, depending on the breaking conditions,
e.g. with more or less particles produced during the cracking. Secondly, and more severe for
the isotopes, the δ13C and δ18O values were shifted by around +1‰ for δ13C and around +2‰
for δ18O. This experiment indicated that an adsorption process with kinetic fractionation happened at the fresh glass surfaces. Although the tube cracker technique has been used for many
decades, until now it was only used for dual-inlet Multiport measurements with CO2 amounts
a factor of 100 larger than in this study. Interestingly, isotopic fractionation connected with
the adsorption of CO2 on fine glass particles were already reported for a dual-inlet application
(Coleman, 1996). Although the amount of glass particles produced from the cracking can be
reduced by scoring the tube, however, due to the tiny dimensions of the tube (1 mm o.d.) this
is not reliably enough. Consequently, the other option is to inhibit the CO2 adsorption. This
was accomplished by providing a strong adsorbent in excess to the He carrier: water vapor.
The He of the GC carrier gas flow is bubbled through deionized water within a ¼’’ o.d. glass
tube, similar to the humidifier used by Leckrone and Hayes (1997). The excess of H2O compared to CO2 (molar ratio H2O/CO2 ~20 within the cracker) then prevents CO2 from adsorbing at the glass surfaces, accordingly, the fractionation phenomenon disappeared after the
Methods and instruments for δ13C and CO2 analysis on ice cores
75
humidifier was installed. At room temperature, the humidifier causes a saturated H2O partial
pressure within the He carrier of 26 mbar.
As a consequence, an extra long Nafion dryer is needed to remove this unusually high load of
water vapor from the He carrier before the stream enters the cryofocus capillary. The Nafion
membrane (0.03’’ o.d. and 50 cm length, Ansyco, Germany) houses in a ⅛’’ o.d. glass tube
and a countercurrent He stream of 5 mL min-1 dries the GC flow (Fig. 3-2).
3.6.6 Cryofocus and gas chromatographic separation of CO2 and N2O
To allow a gas chromatographic separation (Fig. 3-11) of the sample gases, a sharp peak has
to be injected onto the GC column. For this, the CO2/N2O mixture flushed out of the tube
cracker is cryogenically condensed on a cryofocus capillary. A short loop of deactivated fused
silica (0.32 mm i.d., 20 cm length) is used for this purpose, which can be immersed in a LN
Dewar using a pneumatic actuator controlled by the ISODAT software (see Fig. 3-2). The LN
level of the Dewar is maintained constant during the ~10 hour measurement session. To
achieve this, a LN refilling device was built based on the principle used for the cooling systems described in Chapter 3.5.3 and 3.5.5. The LN level within the Dewar is monitored using
a thermocouple (Fig. 3-2). The automatic refilling of the LN Dewar is deactivated for the time
CO 2
m/z 46
m/z 45
m/z 44
ratio 45/44
1.24
3000
1.22
2000
1.2
Ratio 45/44
Intensity [mV]
4000
N2O
1.18
1000
1.16
0
5
10
15
20
Time [s]
25
30
35
40
Figure 3-11 Chromatogram showing the separation of CO2 from N2O at a GC temperature of 70 °C. Due to the
high mixing ratio of CO2/N2O of ~1000, the separated N2O peak is not visible in the intensity plot (left axis),
but it is visible as a bump between 28 and 37 s in the mass ratio 45/44 (right axis) as the isotopic ratio of N2O is
considerably lighter than CO2. The characteristic up-and down swing of the isotopic ratio 45/44 is due to the
chromatographic effect separating not only gas species, but also their isotopes whereby the heavier 13CO2 molecules elute earlier from the GC column than the lighter 12CO2. For the automatic integration of the separated
CO2 peak using the Finnigan software package ISODAT, start end endslope values of the mass trace 44 are
defined. At this example, the peak integration starts at 11 s and ends at 26 s on this arbitrary time axis. Note
that prior to the calculation of the 45/44 ratio, a constant off-set (50 mV for mass 44 and 59 mV for mass 45)
was added to the ion beams to gain a smooth ratio at the edges of the peak when intensities are low.
Methods and instruments for δ13C and CO2 analysis on ice cores
76
the cryofocus is immersed in LN to avoid interference. This is accomplished using a relay
switch controlled by the ISODAT software (detailed description in Chapter 3.7.3).
The sharp sample pulse from the cryofocus is directed to the GC, where CO2 is separated
from N2O and drilling contaminants using a Porabond column (25 m length; 0.32 mm i.d.; 5
µm film thickness; Varian) isothermally at 70 °C (see sample gas chromatogram in Fig. 3-11).
Contrary to the preferred GC temperature of 40 °C, used e.g. by Ferretti et al. (2000), 70 °C
was used for this application. At higher temperatures the GC conditions were found more stable during the long measurement sessions spanning ~10 hours. At lower temperatures the
δ13C values of the reference gas peaks often drifted with time and showed a larger dependency on the peak intensity (‘linearity’ of the CF periphery). Although peak separation of CO2
and N2O is reduced at higher temperatures (for comparison: distance of peak maxima for
CO2-N2O is 25 s for 40 °C, but only 13 s at 70 °C), operation at 70 °C was chosen due to
more stable measurement conditions. The isotopic separation of CO2 strongly depends on the
temperature at which the GC was operated (Fig. 3-12). At 40 °C the chromatographic isotope
effect leads to a fractionation of the CO2 peak ranging from -100‰ to +100‰, which is large
compared to the attained measurement precision of 0.05‰ after peak integration. As visible in
Figure 3-11, the CO2 peak shows a tailing due to adsorption effects of CO2 in the GC system
Ratio 45/44
1.25
1.2
100
50
0
Ratio 45/44 [‰]
150 °C
80 °C
70 °C
40 °C
1.3
1.15
-50
1.1
0
5
10
15
20
25
30
35
Time [s]
Figure 3-12 Comparison of the isotopic separation of CO2 reference gas peaks due to the gas chromatography.
The maximum peak intensity of mass 44 was ~5 V for all experiments at ~16 s on the arbitrary time scale of the
x-axis. For higher temperatures the chromatographic isotope effect during the passage of CO2 through the GC
becomes smaller as the interaction of CO2 with the column becomes smaller as well. The left y-axis shows the
ion beam ratio of mass 45 and mass 44 (note, mass 45 beam is multiplied with 100 due to different resistors of
the Faraday cups). On the right y-axis the 45/44 ratio was transformed to ‰ to visualize the strong chromatographic effect on CO2 ( 45/44 / mean(45/44) -1) x 1000). Prior to the calculation of the 45/44 ratio, a constant
off-set (50 mV for mass 44 and 59 mV for mass 45) was added to the ion beams to gain a smooth ratio at the
edges of the peak when intensities are low.
Methods and instruments for δ13C and CO2 analysis on ice cores
77
as well in the ion source. Due to the peak tailing, the integrated peak area is ~97% and induces a small bias in the δ13C value depending on the selected integration boundaries. With
different peak intensities the integrated fraction varies according to the integration parameter,
i.e. choosing a fixed end slope value results in a slightly earlier cut-off of the peak’s tail, thus,
a slightly smaller percentage of the integrated peak area. Due to this dependency of the peak
integration on the peak intensity, usually a dependence of the δ13C on the peak intensity is
observed (‘linearity’), which has to be corrected for (Chapter 3.8).
Since the isotopic fractionation is largest at 40 °C, adsorption effects should have a higher
impact on the long term measurement stability, which might be responsible for the observed
less reproducible conditions at 40 °C. On can speculate that at a higher GC temperature the
memory effect decreases, i.e. the equilibration of the column is quicker and it saturates at a
lower level. The shape and extent of the linearity effect is slightly changing from one measurement session to the other. Therefore, the system’s behavior has to be determined for each
session to account for altered measurement conditions. The effect of linearity or amount dependency of the δ13C value is a common phenomenon with GC-IRMS applications (MeierAugenstein et al., 1996; Hall et al., 1999; Schmitt et al., 2003).
While the successful GC separation of CO2 from N2O can be visualized in the chromatogram
as a small N2O bump eluting after the large CO2 peak (Fig. 3-11), unfortunately this is not
possible in case of the drill fluid impurities. Since the Porabond GC column was introduced,
no sudden distortion of the 45/44 ratio was observed any longer for ice core samples. As it is
unlikely that all measured samples were free of this contamination, it must be assumed that
these substances do not elute from the column at 70 °C, but are retained on the column. Possibly, these substances are flushed from the column in the ‘stand by mode’ when the GC temperature is set to 200 °C to remove water traces accumulated during the operation at 70 °C.
3.6.7 Open split and IRMS measurement
The open split acts as the interface coupling the pressurized He flow system with the high
vacuum side of the ion source of the mass spectrometer. The GC carrier gas flow enters the
open split of the GP-Interface (Finnigan, Bremen, Germany) and only a fraction actually enters the ion source of the IRMS (see Fig. 3-2 for details). As the open split operates at atmospheric pressure, a fixed flow rate of 0.3 mL min-1 is directed to the ion source, regardless of
pressure and flow rate of the GC carrier. The remainder of the carrier flow is discarded to the
atmosphere, i.e. is lost for the analysis. The split ratio is defined as the ratio of the total carrier
gas flow to the flow directed to the ion source. This ratio usually ranges between 5 and 10
depending on the application’s carrier flow. Applications developed for the δ13C analysis of
large atmospheric samples operate at a high carrier flow, hence, a high sample loss due to
78
Methods and instruments for δ13C and CO2 analysis on ice cores
elevated split ratios, e.g. Ferretti et al. (2000) used 3 mL min-1 and Ribas-Carbo et al. (2002)
1.5 mL min-1. For small ice core samples, a large sample loss at the open split has to be prevented to achieve the required precision of <0.05‰. Leuenberger et al. (2003) used 1.0 mL
min-1 and in this work the flow rate was further reduced to 0.85 mL min-1, resulting in a split
rate of 2.8. This value is at the lower boundary of the proper working range of an open split.
To prevent inward diffusion from the atmosphere, the open split is purged with 1.4 bar He
with the standard setup of the Finnigan GP-Interface.
A Finnigan MAT 252 in continuous flow mode equipped with a universal triple collector having three Faraday cups tuned for the masses 44, 45, 46, is used for the measurement of the
isotopic ratios δ13C and δ18O. The continuous ion currents from the three Faraday cups are
converted to voltages using resistors with 3 x 108 Ω for mass 44, 3 x 1010 Ω for mass 45, and
1 x 1011 Ω for mass 46. An integration time of 0.25 s was selected as sampling rate for the
three mass traces.
3.7 Description on of the analysis procedure
The following section describes the entire analysis procedure in the laboratory for the ice core
samples and the whole air standards. First the sublimation system, in which the air is extracted from the ice sample mostly by manual handling of the valves and the cryogenic traps.
Secondly, the tube cracker-IRMS system, with which the stored sample tubes are transferred
in a Helium flow, cryofocused, purified and finally measured. The latter system is highly
automated, except the tube cracker, using pneumatic valves, which are operated by the Finnigan software ISODAT.
3.7.1 Sublimation extraction for ice core samples
Initialization and sample preparation
The first step of the sublimation extraction procedure is mounting a cleaned glass tube to the
Ultra Torr fitting (see Fig. 3-7). Then V5 is opened and the air pumped to vacuum. To remove
any impurities or water adsorbed on the glass surface, the tube is heated with a small torch.
Prior to the next sample is loaded, the cooling system is started to cool down the sublimation
vessel at temperatures of around -5 °C. To provide a clean surface, each ice cube has to be
prepared separately in the cold lab in the following way: The ice cube is cut with a band saw
to the dimensions 3.3 cm x 3.3 cm x 4.8 cm length. To fit this cube into the sublimation vessel
with an internal diameter of 3.3 cm, the edges are rounded and trimmed with a stainless steel
knife to a cylinder of ~3.2 cm diameter and weighing around 33 g.
Methods and instruments for δ13C and CO2 analysis on ice cores
79
Evacuation and cleaning of ice and vessel
The ice is inserted in the precooled sublimation vessel and the vessel then mounted to the
flange via a copper gasket. With the cooling system a temperature of ~-23 °C is adjusted
within the vessel. The air is pumped out of the vessel until a constant pressure reading is
reached corresponding to the vapor pressure of the ice (pH2O of ~0.7 mbar). The air is
pumped off via V2 and V1 passing the large diameter LN water trap to freeze out any water
vapor (Fig. 3-1). The removal of the water is essential to check the entire system for leaks. If
water were not removed, the pH2O would dominate the total pressure within the vacuum system and small leaks, e.g. at the flange, could not be detected. A two-hour pumping step at a
pH2O of ~0.7 mbar is followed since CO2 sorption effects of the vessel surfaces is a critical
issue and desorption from surfaces is most effective at high pH2O pressures (Zumbrunn et al.,
1982; Güllük et al., 1998). In parallel, a few mm ice sublimates from the sample surface and
further cleans the ice core sample. After around 30 min pumping, the pressure of the high
vacuum system (PVAC) approaches <2 x 10-6 mbar as the temperature of the ice and the cooled
glass vessel are in thermal equilibrium, hence, the sublimation is gradually declining. After
one hour and provided the system is leak free, a final pressure of <3 x 10-7 mbar is achieved in
case of bubble free ice, and ~5 x 10-7 for clathrate ice due to slow decomposition of the clathrates continuously releasing some air. After two hours, a few millimeters of ice have been
removed and the surfaces of the vessel and traps are cleaned.
In between, the cooling device of the external water is started 60 min prior to its actual usage
for the sublimation of the ice sample to reach the set-point of -140 °C±2 °C and allow temperature equilibration within the trap.
Preconditioning sublimation
The sublimation of the ice sample is started by (1) slowly rising the current of the halogen
bulbs, thus providing energy for the sublimation of the ice sample, (2) simultaneously decreasing the temperature of the cold air stream to -120 °C and (3) increasing its flow rate, using the regulator for compressed air and V14 + V15, that a pH2O of ~0.5 mbar results. This
corresponds to an ice temperature of -28 °C being well below where a quasi liquid layer might
form at the ice’s surface. The gas stream liberated from the ice is now directed to the trapping
system by closing V2 and directing the gas flow through the traps (V3-V9 being opened, V8
closed). In order to achieve pressure equilibrium within the system and a saturation of the
surfaces with sample CO2, sublimation is run for ~3 min without collecting the trapped air for
further measurement. This step is called preconditioning. A detailed time table showing the
manual sequence of switching of the valves and cooling and heating of the traps is given in
the Appendix (Fig. 5-1).
Methods and instruments for δ13C and CO2 analysis on ice cores
80
Main sublimation
During the main sublimation the gas release from the ice sample is continuously operating
and in total 5 sub-sample tubes are being collected in a repeated sequence. The details for this
procedure are summarized in Figure 5-2 in the Appendix. Briefly, the released sample gases
are collected for 11-20 min onto the CO2 trap and the molesieve trap, which are both immersed in LN. The respective times for each sub-sample number (corresponding to tube No.
1-5), are listed in Table 3-3. When the trapping duration is terminated, trapping is interrupted
and V3 closed for around 3 min. During this time interval the condensed CO2 is vaporized
from the CO2 trap and finally transferred to the glass tube. This CO2 transfer requires two
stages to achieve 100% recovery: First, from the CO2 trap to the top of the capillary and in the
second step to the tip of the capillary. The second step takes rather long (6 min) to accomplish
a quantitative transfer into the small diameter capillary. In parallel, the molesieve trap is
automatically heated to 100 °C and the air is released to the insulated expansion volume. After pressure equilibrium, the pressure is read out at the PM transducer and the temperature of
the expansion volume is noted to allow a temperature correction. Both traps are then evacuated and again immersed in LN to cool them down to -196 °C, V3 and V4 are opened and the
accumulated gas in the sublimation vessel is passed through the cold traps to begin the next
collection cycle. Note that at this time, the second step of the CO2 transfer within the tube has
not yet been finished. After 6 min, the tube is closed at 2 cm length with a small gas torch.
The whole sequence is repeated 5 times until the ice is almost consumed.
Table 3-3 Shown are the applied process parameters for the five sub-samples or tube numbers. As the ice sample
continuously shrinks during sublimation, the time for sample collection has to be increased to achieve similar
amounts of released air per tube. To prevent excessive sublimation durations, IR emission is successively increased and the cooling capacity adjusted accordingly to operate the sublimation at constant temperature.
sub-sample/
tube
[No.]
1
trapping duration
[min]
11
IR lamp
power
[W]
200-240
amount of air released
[mL STP]
pressure of cooling system
[bar]
~0.4
0.5
2
13
240
~0.45
0.5-0.55
3
15
240-280
~0.5
0.55
4
18
280-330
~0.55
0.55-0.6
5
20
330-400
~0.5
0.6
Methods and instruments for δ13C and CO2 analysis on ice cores
81
Regeneration of the system for the next ice sample
The following routine was applied to reinitialize the sublimation system for the next ice sam-
ple. Basically, the sublimation vessel and the external water trap are rapidly warmed up using
compressed air at ambient temperature to melt the ice for removal.
•
the automatic LN refilling for the cold air system and the external water trap are
switched off
•
an compressed air stream is directed through the external water trap (see Fig. 3-6),
after thawing of the ice (~100 mg) condensed within the external water trap, V3 is
opened and the water transferred to the sublimation vessel
•
compressed air is directed through the cooling jacket of the sublimation vessel
(V14 is closed, while V16 is opened; see Fig. 3-5 for details) thereby the condensed ice is melted and after opening the flange the meltwater is removed and the
vessel dried
•
the molesieve trap is constantly heated to 140 °C for 1 h to remove water traces
3.7.2 Procedure to verify the analysis with air standards
In the following, the preparation of air standards using the reference inlet is described. The
aim is to introduce the air standard into the sublimation vessel and process this air identically
to the air released from an ice core sample. Thus, with these air standards the analysis of ice
core samples is simulated as closely as possible to verify the procedure and to calculate the
total off-set of the analysis. As already stated in Chapter 3.5.8 two different methods were
applied:
a) introduction of air standards with sublimation of blank ice
Generally, this procedure closely follows that for ice core samples described in Chapter 3.7.1
with the two following modifications.
•
Instead of an ice core sample, bubble free ice is inserted into the sublimation vessel, which
is cut and prepared identically to a sample. It was kindly provided by the University of
Bern and was produced by degassing distilled water using a zone melting - refreezing approach (M. Eyer, personal communication 2005). This blank ice (‘Bern blank ice’) is free
of visible air inclusions and expected to be free of CO2.
•
The reference inlet (see Fig. 3-2 for details) has to be preconditioned already 90 min prior
to the start of the sublimation to remove stagnant gas and to equilibrate the surfaces. To do
this, V10 is closed while V11 and V12 are opened. The inlet is flushed five times within
30 min by switching V13 for 10 s from the low purge to the high purge capillary with a
82
Methods and instruments for δ13C and CO2 analysis on ice cores
flow rate of ~140 mL min-1. After repeatedly flushing the inlet at a high flow rate, V13 is
set to the low purge flow to purge the inlet for 60 min at a flow rate of 0.5 mL min-1.
The blank ice is then sublimated and the continuously introduced air standard processed to
tubes as described above in Chapter 3.7.1.
b) introduction of air standard without sublimation of blank ice – ‘static’ conditions
In this alternative version, air standard tubes are processed under ‘static’ conditions, i.e. the
ice in the sublimation vessel is not sublimating, but the ice is held isothermally at the same
temperature and pH2O conditions like during the sublimation. To do this, the description for
version a) is followed, but instead of starting the sublimation, the temperature and the flow
rate of the cold air stream are adjusted to -30 °C and 6 L min-1 to yield a pH2O of ~0.4 mbar
within the sublimation vessel.
Note that bubble free blank ice can be produced also ‘in-situ’ within the sublimation vessel by
refreezing meltwater. If this is done carefully, the water slowly freezes from the bottom to the
top and the freezing front expels any dissolved gases during crystallization without breaking
the glass vessel. Blank ice produced with this refreezing method, combined with the ‘static’
conditions was the preferred procedure to process air standards. In a couple of experiments
‘Bern blank ice’ was used instead, and these air standards will be referred accordingly ‘Bern
blank ice’ air standard.
3.7.3 Tube cracker-GC-IRMS measurement scheme
According to Figure 3-9, a measurement session is started with std on/off pulses generated
with the GP-box reference port (see Fig. 3-2). The proper measurement with the cracker-GC
system begins with a ‘linearity section’ followed by the randomized measurement of the sample tubes to prevent a bias resulting from the measurement order. After the measurement of
sample tubes is completed, a second ‘linearity section’ and std on/off pulses terminate the
session. The detailed time scheme describing the operation of the valves and the movement of
the cryofocus actuator is provided in the Appendix (Fig. 5-3). Briefly a measurement cycle for
a sample tube (EQ-SA) is executed as follows:
The LN Dewar of the cryofocus is automatically refilled until the set-point level is reached.
The tube cracker is first flushed with the high cracker flow. After switching the Valco (6P2)
valve to the ‘transfer mode’, the GC carrier flows through the cracker. The CO2 pulses are
admitted to the cracker by switching the Valco 1 (6P1) from ‘fill’ to ‘inject’ and reverse. Each
time 10 µL with ~3 nmol CO2 are transferred from the reference loop to the GC carrier and
pass the tube cracker. For 90 s, the cryofocus capillary is immersed into LN, CO2 is trapped,
and after lifting the capillary from LN a sharp peak of CO2 enters the GC column. After 30 s
Methods and instruments for δ13C and CO2 analysis on ice cores
83
delay, Valco 2 is switched to the clean mode and the GC carrier bypasses the cracker. To load
the cracker with a new tube, the cracker is opened, the glass shards of the last tube are removed and the next tube is carefully inserted. After closing it, the cracker flow sweeps the
atmospheric air to the vent. Flushing the cracker in this ‘clean-mode’ takes 60 s. Valco 2 is
then switched back to the ‘transfer mode’ and the GC carrier passes the cracker again.
Through bending the cracker, the scored tube is broken into two pieces and the sample gas
transferred to the cryofocus capillary immersed in LN. After cryofocusing the sample for 90 s,
the capillary is automatically lifted and after a delay of 30 s the cycle begins again for the next
EQ reference gas peak.
From this follows that the previous sample peak is still on the GC column, while the cracker
is prepared for the following reference peak (or vice versa). The automatic procedure for the
linearity section (EQ-L1, L2, and L3) is identical to the procedure described above, with the
exception that CO2 pulses are admitted to the cracker instead of cracking a tube. Moreover,
the cracker is not opened in between. The injection of CO2 to the carrier is executed at the
precise time the tube would be cracked in case of the sample procedure. To produce reference
peaks with different peak sizes (L1, L2, L3), the reference loop is filled and flushed several
times accordingly. Note that due to the cryofocus the general peak shape for different peak
sizes is identical for tube samples and reference peaks, which is a prerequisite for applying the
identical treatment principle.
Following this alternating EQ-SA scheme, all sample tubes are measured. A final ‘linearity
section’, to check whether the system’s linearity changed during the analysis, completes the
IRMS run. A measurement session is ended with a series of std on/off peaks to check the performance of the IRMS. Having both the information from std on/off and the linearity peaks
admitted to the cracker system one can distinguish whether an observed drift phenomena is
caused by the mass spectrometer or from the cracker system.
3.8 Raw data processing and performance of the CF-IRMS analysis
The following paragraph describes calculations applied on the raw data of the IRMS results to
correct the measured tubes of ice core samples (SAice) and air standards (SAair) for instrument
drift and linearity effects. As stated above, a set of 20-30 tubes with extracted ice samples and
air standards was analyzed within a single session (see Fig. 3-9). The calculated correction
functions for drift and the linearity effect may slightly change from one session to another.
Note that additional corrections for ice core samples (air standard off-set and correction for
gravitational settling) are described in Chapter 3.9 and 4.4. The entire step by step correction
scheme for ice core samples is given in Figure 5-4 in the Appendix.
Methods and instruments for δ13C and CO2 analysis on ice cores
84
3.8.1 δ13C
3.8.1.1 Corrections by the IRMS software
To derive internationally referenced δ13C values from the three ion current signals m44, m45
and m46, four basic steps are necessary:
•
definition of the integration boundaries (start and end slope of the peak)
•
background correction of the ion currents
•
δ17O correction
•
shift of the working standard to the VPDB scale
These steps were performed using the Finnigan software package ISODAT NT, with which
the CF-IRMS system was operated. As shown above (Fig. 3-11), the CO2 peak entering the
ion source is almost of Gaussian shape and N2O resides on the tail of the much larger CO2
peak. The side effect of the GC separation is that the isotopic composition within the CO2
peak is highly fractionated (Fig. 3-12). While the front is enriched with the heavier isotopologues, the tail is isotopically depleted. Consequently, the maxima of the three mass traces
are shifted on the order of 0.2 s. Usually, a so called time-shift correction is made to align the
maxima of the three traces prior to integration. Unfortunately, the time-shift correction embedded in the ISODAT software produced noisy results for the selected integration time of
0.25 s, which might be due to the coincidence that the time shift and the integration time have
quite similar numbers. A shorter integration time, whereby increasing the number of data
points, cannot be applied for long measurement sessions due to the software’s limited storage
capacity. Therefore, the time-shift correction was disabled. However, this can be justified by
the fact that the off-set is rather small, and as reference gas peaks and sample peaks were
treated identically, the off-set cancels out.
To define the integration boundaries for each peak, a threshold for the start slope of 1 mV s-1,
and 20 mV s-1 for the end slope was found to gain most reproducible results. A background
signal of mass 44 of 2-3 mV was subtracted using the ‘individual background’ setting. Before
δ13C values are calculated from the integrated ion currents, the contribution of the oxygen
isotope
17
O to the mass 45 has to be quantified, the so called ‘Craig correction’, (Craig,
1957). This correction is necessary since not only 13CO2 contributes to mass 45, but also the
rarest of three oxygen isotopes, 17O. For this, the detection of three masses (44, 45, 46) is required to determine the
contribution of
17
13
C/12C ratio of CO2. Prior to the calculation of the δ13C value, the
O to mass 45 is calculated from δ18O using mass 46 and the natural abun-
dance ratio between δ17O and δ18O of 0.5146; for details see review by Santrock et al. (1985).
These δ13C values are initially measured against the isotopic composition of an internal reference gas or working standard. To report these values relative to the international VPDB scale
the following equation was used:
Methods and instruments for δ13C and CO2 analysis on ice cores
δ (SA-VPDB )
= δ (SA-Ref ) + δ ( Ref-VPDB ) +
85
δ (SA-Ref ) ⋅ δ ( Ref-VPDB )
(3 − 1)
1000
where δ(SA-VPDB) denotes the δ value of the sample (SA) relative to the VPDB standard, δ(SA-Ref)
the δ value of the sample to the internal reference (Ref) ,and δ(Ref-VPDB) the δ value of the internal reference gas against the VPDB standard. After these basic corrections, the data was exported to MATLAB where the data set of each measurement session was further processed.
3.8.1.2 Drift with time
A frequently observed feature during IRMS measurements is that the isotopic ratio is not stable, but rather drifts with time. To correct for this drift and identify its origin during the measurement, CO2 reference gas of known isotopic composition was directed to the IRMS. During
a measurement session two different types of standards were used: First, std on/off pulses,
which do not pass the tube cracker system, but are directly introduced to the ion source. Secondly, reference peaks, which pass the entire tube cracker system (EQ peaks and linearity
peaks L1, L2, and L3 with different intensities, shown in Fig. 3-9). While the std on/off pulses
provide information only at two points (start and end of the measurement), a continuous time
series spanning the 10 h measurement is available in case of the EQ peaks (Fig. 3-13). From
this time series a slope is calculated to detrend the δ13C values for the entire measurement.
Although less robust due to only two tie points, a statistically identical slope is calculated
−6.7
EQ raw data
EQ linear trend
std on/off
std on/off linear trend
−6.75
−6.8
13
δ C [‰]
−6.85
−6.9
−6.95
−7
−7.05
−7.1
−7.15
0
1
2
3
4
5
Time [h]
6
7
8
9
10
Figure 3-13 Comparison of the δ13C drift observed during a measurement session of 10 hours. In this example
the δ13C values of EQ reference pulses (blue dots) linearly increase with time with a slope of 0.012‰ h-1 (blue
line). The std on/off pulses directly introduced into the ion source of the IRMS at the start and the end of the
measurement session show almost the same slope 0.014‰ h-1 (black dotted line). After the δ13C values were
detrended the standard deviation of the EQ peaks is 0.04‰. The observed off-set between std on/off and EQ
measurements of ~0.07‰ is rather small given the fact that the CO2 reference gas is treated completely different prior to injection to the ion source.
86
Methods and instruments for δ13C and CO2 analysis on ice cores
from the std on/off measurements. As visible in Figure 3-13, for this measurement both slopes
were almost identical indicating that the observed drift is not due to the tube cracker system,
but caused by the IRMS instrument alone.
3.8.1.3 Amount dependence of δ13C values – ‘linearity’
A second feature encountered in IRMS applications is that the isotopic ratio slightly depends
on the peak intensity, i.e. the amount of gas introduced into the ion source. Within the isotope
community this behavior is generally termed ‘linearity’, though the relation between δ13C and
the peak intensity is rarely a linear function, but often exponential (Hall et al., 1999). To correct for this effect, reference peaks with three different intensities (‘linearity’ peaks) were
admitted to the tube cracker (L1, L2, L3). Note that std on/off peaks cannot be used for this
purpose as the std on/off procedure generates roughly a rectangular peak shape, whereas
peaks transmitted through the cracker system have a Gaussian shape. Further, the latter peaks
have the characteristic up and down of the 45/44 mass ratio due to isotopic fractionation induced from the GC separation (see Fig. 3-11). As both the peak shape and the isotopic fractionation within the peak influence peak integration and the resulting δ13C value, sample and
reference peaks must be identically in this respect (Meier-Augenstein et al., 1996).
The left plot in Figure 3-14 illustrates the linearity effect of a measurement session with the
tube cracker-GC-IRMS system. As shown above (Fig. 3-9), a section of linearity peaks is
measured before and after the sample tubes, in total 9 peaks per size class. Their amplitudes
range from ~1.7 V for L1 to ~5.1 V for the L3 peak, with the δ13C value nonlinearly decreasing with increasing amplitude (-6.82‰ for 1.7 V, - 6.90‰ for 3.4 V, and -6.93‰ for 5.1 V).
The data is best described with a first order polynomial fit of the logarithm of the peak amplitude (intensity of mass 44 = i44) marked with a blue line in Figure 3-14.
The δ13C values of reference peaks and tube peaks were corrected for this linearity effect with
Eqn. 3-1 as described previously (Hall et al., 1999; Schmitt et al., 2003):
δ 13 C corr = δ 13 C raw − m ⋅ log( i44 ) + c
(3 − 2 )
with δ13Ccorr the corrected δ13C values, δ13Craw the detrended δ13C values, m and c the coefficients of the polynomial, and i44 the peak intensity (maximum amplitude in mV).
The corrected δ13C values of the linearity peaks are shown in Figure 3-14 on the right plot.
The precision of the reference peaks admitted to the cracker system and measured with the
IRMS is on average 0.05‰ for the small peaks, and 0.03-0.04‰ for the larger peaks. According to theoretical assumptions of the ion current measurement in the IRMS instrument, the
attained precision with the whole CF-IRMS system is about 2 times the theoretical shot noise
limit calculated after Merritt and Hayes (1994).
Methods and instruments for δ13C and CO2 analysis on ice cores
87
Note that the amplitudes of sample tubes generally fall within a narrow range from 3 to 5 V
for Holocene ice. As the linearity effect for high amplitudes is only small, the absolute δ13C
correction is rather small too and amounts on average ~0.01-0.02‰. For ice with low CO2
concentrations, like from the glacial period, the absolute correction is ~0.03‰ as the sample
amplitudes fall in the steeper range of the linearity function. As the coefficient of determination of the linear regression line with Eqn. 3-1 is on average ~0.7, the error introduced by the
linearity correction is <0.01‰ for all samples.
−6.6
−6.82
−6.91
−6.93
−6.6
−6.93(0.05)
2000
3000
4000
Amplitude [mV]
−6.92(0.03)
−6.7
δ13C [‰]
δ13C [‰]
−6.7
−6.92(0.05)
−6.8
−6.9
−7
−6.8
−6.9
−7
linearity peaks (L1, L2, L3)
polynomial 1.order log(x)
−7.1
2000
3000
4000
Amplitude [mV]
−7.1
5000
5000
Figure 3-14 Left plot: dependence of the δ13C value on the peak amplitude or signal intensity (‘linearity effect’), numbers indicate the δ13C means for the three peak classes (L1, L2, L3); the blue line illustrates the
polynomial fit of the logarithm of the peak amplitude. Right plot: corrected δ13C data with values for means
and standard deviations (1σ) given on top.
3.8.2 CO2 concentration
To calculate CO2 concentrations for the extracted air of the ice samples, a similar approach as
used by Ribas-Carbo et al. (2002) and Eyer (2004) was applied using the following equation:
CO2 ( sample) =
isample ⋅ pair − std
psample ⋅ iair − std
⋅ CO 2 (air − std )
(3 − 3)
with CO2(sample) and CO2(air-std) denoting the respective CO2 concentrations in ppmv for
the ice core sample and the air standard, isample and iair-std the respective ion current intensities
of mass 44 (peak maximum) for the sample and air standard; psample and pair-std the pressure
reading of the expanded air measured with the pressure transducer PM at the molesieve trap.
Note that prior to the calculation, the raw pressure reading from the transducer was corrected
Methods and instruments for δ13C and CO2 analysis on ice cores
88
with the temperature reading, Tex (the thermocouple attached to the expansion volume in Fig.
3-1), for the influence of temperature variations using the ideal gas law.
For each measurement session 5-10 air standard tubes were prepared together with 15-20 ice
sample tubes. To achieve a robust calibration, the pair-std/iair-std ratio for each standard tube was
calculated and after removing outliers the mean pair-std/iair-std was then used in Eqn. 3-1 to calculate CO2 concentrations for the samples.
Contrary to δ13C, where the systematic trend of the EQ peaks can be used to correct the raw
δ13C values of the sample tubes, the intensity of the EQ peaks cannot be used to correct the
sample tubes in case of the CO2 concentration. Although for some measurement sessions a
trend was visible in the m44 intensity time series of the EQ peaks (see example in Fig. 3-15
left panel), no relation was found between the intensity of EQ peaks and the time series of the
calculated CO2 concentration of sample tubes. It can be reasoned that the scatter in the intensity of the EQ peaks does not reflect changes in the performance of the cracker-GC-IRMS
system. Rather, the intensity changes of the EQ peaks probably reflect small CO2/He mixing
ratio fluctuations within the reference gas device (Fig. 3-2). Note that an observed trend in the
m44 intensity does not necessarily correspond with δ13C fluctuations also indicating two independent causes for both parameters (Fig. 3-15 right panel).
The relative standard deviation (standard deviation divided by its mean) for EQ peaks is
~0.4%, this translates to a standard deviation of ~1.1 ppmv assuming a sample with a CO2
concentration of 280 ppmv. Since an additional factor independently contributes variability to
the EQ intensity without adding variability to the sample measurement, the 1.1 ppmv standard
deviation can be regarded as an upper estimation for the measurement error of the CO2 concentration measurement. The second source of error for the CO2 concentration (according to
Eqn. 3-2) is due to the error introduced from the pressure reading (reading precision ±0.0001
mbar for an absolute value of ~0.5 mbar) including the uncertainty from the temperature
−6.8
5200
raw δ C [‰]
5150
5100
5050
5000
4950
−6.9
13
Amplitude [mV]
5250
−7
−7.1
0
2
4
6
Time [h]
8
10
4950 5000 5050 5100 5150 5200 5250
Amplitude [mV]
Figure 3-15 Left: time series of the m44 amplitude of the EQ peaks. The right panel shows the scatter of the δ13C
raw data with amplitude for the EQ peaks.
Methods and instruments for δ13C and CO2 analysis on ice cores
89
(reading precision ±0.1 K at 293 K). The uncertainty introduced from the reading precision is
0.1 ppmv for both parameters.
As the CO2 concentration of the air standard is 277.7 ppmv, thus, close to average CO2 concentration of the Holocene, this 1-point calibration approach of Eqn. 3-1 is justified to gain
accurate values at least for the Holocene period. For the considerably lower glacial CO2 concentrations a second air standard would be required to check whether the relation in Eqn. 3-1
still holds for low CO2 concentrations. As the main focus of this study was δ13C and its small
scale variability, the accuracy of the absolute CO2 concentration had second priority so far.
However, more important are the relative differences among adjoining samples and among
the five sub-samples collected during the sublimation extraction of one ice sample as those
allow clues about small scale variability within the ice itself and help identify processes during the sublimation extraction.
3.9 Results from air standard and blank measurements
To characterize the performance of a newly developed analytical method usually three parameters are examined: (1) the reproducibility of repeated measurements of standards, (2) the
absolute accuracy or deviation of the measured value from the expected ‘true value’, (3) the
amount of analytical blank when no sample is processed. The following section summarizes
the results of air standards admitted to the sublimation apparatus and discusses blank measurements. As pointed out above, air standards mimic the sample treatment as closely as possible and therefore serve as the reference on which the δ13C values and CO2 concentrations of
ice core samples are based on. Air standards providing the reference base were prepared using
the ‘static’ setting (Chapter 3.7.2) and were measured jointly with a set of samples. As these
standards were processed and measured over a period of 4 months, this data set is used here
(Chapter 3.9.1 and 3.9.2) to calculate the analysis reproducibility to which the reproducibility
of the ice core measurements can be compared.
Besides these regular air standards, further experiments were conducted to investigate the
influence of different settings during the admission of air standards (admission of air to ‘blank
ice’ with and without sublimation and a variation of the flow rate). As illustrated in the left
panel (Fig. 3-16) the rate with which air is released from ice core samples during the sublimation is not constant as it is for air standards. Therefore, to identify and estimate ‘side effects’
due to different air release rates between ice core samples and air standards a number of experiments were conducted. These results are discussed in Chapter 3.9.3.
Methods and instruments for δ13C and CO2 analysis on ice cores
0.6
0.04
Air volume [mL STP]
Flow rate [mL STP min
−1
]
90
0.03
0.02
0.01
0
ice core samples
admitted air standard
1
2
3
Tube No.
4
5
0.4
0.2
ice core samples
admitted air standard
0
1
2
3
Tube No.
4
5
Figure 3-16 Comparison of procedural parameters between air standards (red) and ice core samples (diamonds,
EDML 420 m). For the regular procedure to process air standards (‘static’ conditions), the flow rate was kept
constant at ~0.04 mL min-1 (left panel). As the collection time for air standards was fixed at 14 min, a constant
air volume of 0.57 mL STP results for all five tube numbers (right panel). This is slightly higher than the volume collected from Holocene ice core samples. As the ice sample is continuously shrinking during the ongoing
sublimation the collection time is gradually increased (from 11 to 20 min) to compensate the reduced air release
rate and gain comparable total air volumes (see Fig. 3-3).
3.9.1 Reproducibility of δ13C and overall accuracy for air standards
Before the δ13C data are analyzed for reproducibility and overall accuracy, a first step is to
detect differences among the five tube numbers. To this end, the calculated off-set to the assigned δ13C value of the air standard (Tab. 3-4) was subtracted from each data set to remove
systematic differences among individual measurement sessions. Afterwards, the data sets
were combined and the mean and standard deviation (1 σ) calculated for each tube number
(Fig. 3-17, diamonds with error bars). If no differences were present among the five tube
numbers, the means would be identical and the standard deviation zero. However, slight but
−2.5
−2.6
δ13C [‰]
−2.7
−2.8
−2.9
mean of tube No. + − 1 σ
mean of air standard
mean + − 1 σ
−3
−3.1
1
2
3
Tube No.
4
5
Figure 3-17 Compilation of δ13C results of air standard measurements according to the tube number. The data
was compiled from five measurement sessions (total n=33, after removing 2 outliers).
Methods and instruments for δ13C and CO2 analysis on ice cores
91
not significant differences can be observed in Figure 3-17. The red solid line marks the assigned δ13C value to which the averaged standard deviation of the seven data sets for tube
numbers 1-5 (average σ ± 0.07‰ from Tab. 3-4) was added (dotted red lines). Though not a
statistical outlier, tube number 1 is on average 0.05‰ more negative than the other four tubes.
More pronounced is this anomaly for the corresponding CO2 concentration as visible in Figure 3-18. As the CO2 concentration is lower for tube number 1, a contamination can be readily
excluded. More likely is a loss process associated with the transfer into the very tip of the
glass capillary; for tube number 1, the glass capillary is longest, hence, diffusion into its tip
takes longer to achieve 100% than the following tube numbers as the capillary is getting
shorter. Another possibility for the deviation of tube number 1 would be an isotopic fractionation associated with a loss due to adsorption on unsaturated surfaces.
From this first data analysis it can be concluded that the five tubes do not show considerable
differences among the tube numbers. For the second step, the means and standard deviations
from each data set were calculated over all five tube numbers to yield a measure to describe
the reproducibility and accuracy of the method. Note that for this calculation the off-set to the
assigned value was not subtracted since this constitutes the overall accuracy of the analysis
technique.
The results are summarized in Table 3-4. The standard deviation (1σ) ranged between 0.05
and 0.08‰ with an average of 0.07‰. When tube number 1 is removed, i.e. number 2-5 only,
the standard deviation slightly improves (0.06‰). To characterize the performance of the sublimation extraction, one can compare the reproducibility of these air standards with the reproducibility of the CO2 reference gas pulses of similar peak size. As shown earlier (Chapter
3.8.1), the measurement error of the cracker-GC-IRMS system alone was 0.04-0.05‰ for
pure CO2 peaks. Hence, only little variance was added from the additional steps associated
with the introduction of the air standard to the sublimation extraction apparatus and separation
Table 3-4 Reproducibility of air standards processed with the ‘static’ procedure. For each session the standard
deviation (1σ), the mean over all tube numbers (tube No. 1-5) and the off-set of the mean to the assigned value
of the air standard (-2.75‰) was calculated. In two cases (‘std only’ and ‘1056’) an outlier has been removed. *
For the calculation of the ‘total average’ from the five standard deviations the pooled standard deviation (σpooled)
was calculated according to Eqn. 3-4 (see explanation below).
measurement session:
‘std only’
‘276’
‘253’
‘420’
‘1056’
total
date:
2.12.05
20.12.05
26.01.06
6.2.06
21.2.06
average*
9
5
10
4
5
standard deviation [‰]
0.07
0.06
0.08
0.05
0.07
0.07
mean [‰]
-3.08
-3.06
-3.09
-3.02
-2.93
-3.04
off-set [‰]
-0.33
-0.31
-0.34
-0.27
-0.18
-0.29
number of standards
92
Methods and instruments for δ13C and CO2 analysis on ice cores
of CO2 from the bulk air components. The achieved performance for measuring δ13C on air
standards of 0.06‰ is undoubtedly inferior to the state of the art measurements on large atmospheric samples of typically 0.03‰. However, compared to the results from previous studies on small ice core samples (0.12‰ reported by Eyer, 2004), the performance is very good
and precise enough to resolve the proposed natural variance (~0.5-1‰ on glacial/interglacial
time scales).
Since the numbers of replicates for each measurement session ranges from four to ten, additionally the pooled standard deviation, σpooled, was calculated to exclude the bias from those
measurements with only few replicates. With σpooled being the square root of summed squared
deviations of replicates δ i from their respective means divided by the degrees of freedom, i.e.
the number of samples n minus the number of reported means m.
n, m
∑
σ pooled =
(δ i − δ j ) 2
i , j =1
n−m
(3 − 4 )
Concerning the overall accuracy of the method, the off-set of the measured δ13C to the assigned value was ~0.3‰, except for the last measurement session (‘1056’) with an off-set of
only 0.18‰. This smaller off-set might be due to an observed pressure drop near the end of
the live time of the high pressure cylinder of the CO2 reference gas of the IRMS system.
Therefore, it is likely that not the δ13C value of the air standard changed, but rather the reference basis. Similar shifts noticed Eyer (2004), who speculated that due to the pressure drop
the liquid CO2 phase, which is usually found in the cylinder disappears. With the phase
change an isotopic fractionation is involved shifting the CO2 reference gas. As the δ13C values
of the ice core measurements were finally referenced on the air standard, an absolute shift of
the CO2 reference gas is not critical for the ice core samples’ absolute values.
3.9.2 Reproducibility of the CO2 concentration for air standards
In analogy to the δ13C data analysis presented above, first the differences among the five tube
numbers are discussed. The calculations were identical with the exception that considering an
off-set is dispensable as the value for the CO2 concentration of the air standard was assigned
to the calculated IRMS response (see Chapter 3.8.2). The results are shown in Figure 3-18
with the mean and standard deviation for each tube number class marked with diamonds and
error bars (1 σ). Visible is the distinct deviation of tube number 1, with the mean CO2 concentration being 2 ppmv lower than the overall average for all tubes (red solid line). As discussed
above for δ13C, tube number 1 is probably subject to a systematic loss process, and therefore,
Methods and instruments for δ13C and CO2 analysis on ice cores
93
280
278
276
274
2
CO concentration [ppmv]
282
mean of tube No. + − 1 σ
mean of air standard
mean + − 1 σ
272
270
1
2
3
Tube No.
4
5
Figure 3-18 Compilation of the CO2 concentration of whole air standards processed using the ‘static’ method.
The data was compiled from five measurement sessions (total n=33, after removing two outliers). For each tube
number the mean and standard deviation (1 σ) was calculated (diamonds with error bars). The red solid line is
the assigned value of the CO2 concentration of the air standard. The two red dotted lines represent the averaged
standard deviation of the seven data sets for tube numbers 1-5 (average σ ± 1.5 ppmv). Except for tube No. 1
the means of tube No. 2-4 do not deviate significantly from each other.
treated with caution for the calculation and interpretation of the ice core data. However, as the
missing 2 ppmv of tube number 1 is a stable feature at constant conditions, this data was not
immediately discarded.
The reproducibility for CO2 concentrations of air standards is summarized in Table 3-5. The
pooled standard deviation over all tube numbers was 1.5 ppmv, and 1.4 ppmv for tube numbers 2-5. This performance is comparable to other studies measuring CO2 concentrations on
ice core samples (e.g. 1.7 ppmv reported by Siegenthaler et al., 2005a). However, that the
actual precision for samples is supposed to be lower as constant process parameters (flow rate,
total air amount) were applied for these air standards. As will be shown below, a deviation
from these constant settings introduces an additional uncertainty for the ice core results since
the gas release rates during the sublimation of ice do slightly vary.
Table 3-5 Results for CO2 concentration from five measurement sessions showing the reproducibility of air
standards processed with the standard procedure (‘static’ method). For each session, the standard deviation (1σ)
was first calculated including all data (tube numbers 1-5), and secondly after tube number 1 was discarded from
the data since the first tube number deviated from the others (Fig. 3-18). The pooled standard deviation, or
average reproducibility of the measurement is 1.5 ppmv including all data, and 1.4 ppmv for tube numbers 2-5.
measurement session:
date:
‘std only’
‘276’
‘253’
‘420’
‘1056’
2.12.05
20.12.05
26.01.06
6.2.06
21.2.06
9
4
10
4
5
1.6
1.0
1.5
0.6
1.9
8
4
8
3
4
0.8
1.0
1.6
0.2
0.9
σpooled
tube No. 1-5
number of standards
standard deviation [ppmv]
1.5
tube No. 2-5
number of standards
standard deviation [ppmv]
1.4
94
Methods and instruments for δ13C and CO2 analysis on ice cores
3.9.3 Estimation of ‘side effects’ from air standards
Besides the above discussed air standards processed under ‘static’ conditions, a couple of additional experiments were conducted at modified conditions. First, air standard was admitted
during the sublimation of ‘Bern blank ice’ for a comparison with the ‘static method’ (see
Chapter 3.7.2 for details). Secondly, the rate with which the air standard was admitted to the
sublimation vessel was varied to evaluate its influence on δ13C or the CO2 concentration.
Comparison between air standards processed under ‘static’ and sublimation conditions
Results from three experiments comparing both methods showed that δ13C values were
slightly more negative for the experiments with sublimation of ‘Bern blank ice’. On average,
the latter values were 0.1-0.15‰ more negative and were less precise (1 σ ~0.08‰) than the
‘static’ ones. This discrepancy between both methods was already observed during the early
development stage of the sublimation apparatus (see Chapter 3.5) and led to the conclusion
that experiments conducted with sublimation of ‘blank ice’ were less conclusive than the
‘static’ version. As a result, most experiments further on were conducted using the more reliable ‘static’ method. The speculation is that the ‘blank ice’ might contain traces of atmospheric CO2 that were not removed during the preparation. As the used air standard is relatively heavy (δ13C = -2.75‰), a small amount of isotopically depleted CO2 might be responsible for this shift. The precision of the early measurement set-up did not allow a detection of
this small extra CO2. With the final analysis set-up, a few experiments with sublimation of
blank ice were repeated and resulted in ~3 ppmv higher CO2 concentrations compared to the
‘static’ method without sublimation. Slightly higher CO2 values together with ~0.1‰ more
negative δ13C values points to a contamination of the blank ice. Using an isotopic mass balance, a δ13C value of -12‰ (conceivable for laboratory air) would match the 0.1‰ shift of the
mixture and a 3 ppmv concentration increase. Though the admission of air standard during the
sublimation of blank ice mimics the air release from ice samples more realistically, the δ13C
results from the ‘static’ procedure are regarded more accurate. Until truly CO2-free blank ice
is available, it is more conservative to correct the ice core samples with a system off-set from
the ‘static’ procedure. Note that this ~0.1-0.15‰ off-set uncertainty in the absolute δ13C correction value and ~3 ppmv for CO2 does not compromise the outcome concerning the smallscale heterogeneity in the ice (Chapter 4.3) as the off-set is a constant factor. However, this
off-set is a critical issue for the comparison with other ice core data (Chapter 4.4).
Influence of the admission rate
As shown above (Fig. 3-16), the admission rate of air for the preparation of air standards was
slightly higher than the mean air release rate observed during the sublimation of Holocene ice
core samples. Further, the release rate for ice core samples (Fig. 3-16 and 3-19) shows a sys-
Methods and instruments for δ13C and CO2 analysis on ice cores
95
tematic variation during the sublimation as the ice cube continuously shrinks in parallel to the
increase in supplied infra red energy. However, due to technical constraints in the reference
inlet, a simultaneous variation of the admission rate for air standards (i.e. the pressure within
the reference inlet) during the experiment was not possible so far. This is mainly due to the
large dead volume within the pressure regulator making a quick down-regulation of the pressure difficult. For this, the admission rate was reduced from 0.04 to 0.027 mL STP min-1 prior
to the experiment and then held constant (Fig. 3-19). This reduced rate was chosen to account
for the conditions (in terms of the CO2 flow rate) of glacial ice core samples with low CO2
concentrations. Considering only the flow rate of air (left panel, red line) and the total air volume (center, red dots) admitted with the high admission rate, the corresponding values for
glacial samples (marked as diamonds) are well matched. However, in case of the amount of
CO2 (and also its flow rate) the low admission rate (right panel, blue dots) is probably more
representative for ice core samples with low CO2 concentrations.
Within the uncertainty of the measurement, the means in δ13C for both rates did not differ
(Fig. 3-20). Further, for the low rate even more negative values (~0.1‰) were found for tube
number 1 compared to the high rate (Fig. 3-17). This confirms the notion that the first tube
number has a systematic bias towards lower values. Note that for this experiment the collection time was identical for both admission rates. Consequently, less CO2 was collected in case
of the lower admission rate. This resembles roughly the conditions of the CO2 flow rate for
glacial ice and the experiment was aimed to look for a systematic bias between Holocene and
glacial samples due to the contrasting CO2 concentrations. Although this flow rate change
experiments are suitable to simulate the variation of air release from ice core samples during
0.03
0.02
0.2
1056 m (Glacial)
high rate
low rate
0.01
0
0.4
1
2
3
4
Tube No.
5
1056 m (Glacial)
high rate
low rate
0
1
2
3
4
Tube No.
5
amount of CO2 [nmol]
0.04
Air volume [mL STP]
Flow rate [mL STP min
−1
]
8
0.6
6
4
2
0
1056 m (Glacial)
high rate
low rate
1
2
3
4
Tube No.
5
Figure 3-19 Comparison of process parameters for the sublimation apparatus between the admission of air standard
at a high and at a lower rate, marked with red and blue in each panels, and the corresponding parameters for a ice
core sample from the glacial period. Note that the air content of glacial ice is slightly higher so the collected air
volume almost matches that of the standard (see Fig. 3-16 for comparison).
Methods and instruments for δ13C and CO2 analysis on ice cores
96
the sublimation, it is only an approximation to simulate the different CO2 concentrations between Holocene and glacial periods. Nevertheless, the comparison showed that the flow rate
does not influence the δ13C values. To gain more confidence, air standards with contrasting
CO2 levels will be used in the future to better address this issue.
For the CO2 concentration a difference was observed between both rates. For the low rate, the
CO2 concentrations were systematically lower by 5 ppmv (Fig. 3-20). Here as well, tube
number 1 is shifted towards lower values by ~2 ppmv. The slightly lower values cannot be
explained with an IRMS measurement linearity effect comparable to δ13C. A calculation of
the specific IRMS response for the three linearity peaks, L1, L2, and L3, corresponding to
CO2 amounts of ~3, 5, and 7 nmol, respectively, did exclude the CF-IRMS measurement system to cause this deviation. The specific CF-IRMS response derived from linearity peaks
rather suggests CO2 values to be ~1 ppmv higher for the 5 nmol CO2 level (L2) compared to 7
nmol (L3). The latter calculation assumes that the CO2 amount admitted to the cracker by
consecutively filling the reference loop once, twice, and three times precisely scales with the
peak amplitude.
An initial explanation for the observed 5 ppmv difference might be that a constant CO2
amount is lost within the sublimation apparatus itself or during the transfer into the glass capillary, which results in a relative loss for the concentration. A second explanation might be
that the apparent loss in CO2 is due to a non-linearity effect of the air volume measurement
with the pressure transducer. Again, further experiments using air standards with different
CO2 concentrations will help to clarify this effect and reduce the uncertainty of the CO2 concentration measurement.
CO 2 concentration [ppmv]
-2.8
13
δ C [‰]
-3
-3.2
-3.4
-3.6
high rate
low rate
-3.8
-4
1
2
3
Tube No.
4
5
280
270
260
250
240
high rate
low rate
1
2
3
Tube No.
4
5
Figure 3-20 Experimental results showing the influence of the admission rate (high rate 0.04 mL STP min-1,
low rate 0.027 mL STP min-1) on the δ13C values (left panel) and the CO2 concentration (right panel).
Methods and instruments for δ13C and CO2 analysis on ice cores
97
3.9.4 Procedural blanks
Blank experiments were conducted to estimate the proportion of CO2 that either originates
from an external contamination, or from a memory effect. Using an extended vacuum system
like in this work, a steady inflow of air from the outside due to a leakage is the most conceivable way to produce blank values. Yet, air leakages within the vacuum apparatus can be detected from a pressure increase in the high vacuum system (PVac pressure sensor). As copper
gaskets seals were used to connect the sublimation vessel with the vacuum line and the fittings were all-metal, the permeation of air into the vacuum apparatus was small. From the
pressure reading of the vacuum sensor PVac, the leakage rate of air into the sublimation apparatus was <0.001 mL STP h-1, corresponding to <1 ppmv blank assuming 6 nmol CO2 as a
typical sample amount, 400 ppmv CO2 for the ambient air, and a collection time of 15 min.
More critical than the contamination from outside is the desorption of CO2 from the internal
surfaces of the apparatus itself. In gas analytics, CO2 is known as a ‘sticky’ gas species as it
readily adsorbs on metal, glass, and polymer materials to desorb from these surfaces at a later
time. This ad- and desorption behavior was already noticed since the beginning of ice core
CO2 analysis by Zumbrunn et al. (1982) and the problem was constantly confirmed later on
(Güllük et al., 1998; Siegenthaler, 2006). The actual desorption rate from a surface depends
on the surface’s initial CO2 saturation when the surface was exposed to a certain CO2 partial
pressure, on the time constant of the desorption, and its temperature dependency. Further, the
desorption rate depends on the concentration of competing molecules, like H2O, which preferentially displaces CO2 from a surface due to its strong dipole moment. Owing to this complex ad- and desorption processes, a couple of representative blank experiments were conducted to achieve first order estimations for the conditions relevant during the sample preparation and measurement.
Description of blank experiments
Five different blank experiments were conducted to estimate the cumulative contribution of
non-sample CO2 during the analysis process, termed blank contribution:
(1) ‘He-blank’ to estimate the contribution from the He carrier gas. The carrier passing the
tube cracker was cryofocused for 90 s and any collected CO2 afterwards transferred to the
GC. To exclude a memory effect from the cracker, i.e. CO2 desorption from its surfaces, no
CO2 reference gas pulse was admitted to the cracker for one hour prior to this experiment.
(2) ‘cracker-blank’ to estimate the CO2 amount added during the normal operation of the tube
cracker: similar to experiment (1), but preceded by a large CO2 reference pulse (EQ peak).
This procedure estimates the cracker’s memory effect under typical measurement conditions.
98
Methods and instruments for δ13C and CO2 analysis on ice cores
(3) ‘empty tube blank’ to estimate the CO2 release during the flame sealing of the glass tube.
New glass capillaries were prepared identically to the sample procedure (Chapter 3.7.1). The
CO2 trap and the glass capillary were evacuated, V5 closed, and the tip of the capillary immersed in LN. Then the tube was flame-sealed.
(4) ‘CO2 trap blank’ to estimate leakage or outgassing of the CO2 trap. The CO2 trap was held
at LN temperature for 10 min with V4 and V6 being closed, then the trap was warmed up and
the vaporized CO2 transferred to the glass capillary and the tube flame sealed.
(5) ‘sublimation blank’ to estimate the effect during the sublimation of the blank ice. The experiment is identical to the procedure described for air standards (Chapter 3.7.2), but without
admitting air to the sublimation vessel.
Results from blank experiments
With each analytical step, the blank progressively increases as shown in Table 3-6, with a
total sum of 0.03 nmol CO2 measured for the ‘sublimation blank’. The steps with the largest
CO2 contribution are the ‘cracker blank’ and the ‘sublimation blank’. The amount added to
the ‘cracker blank’ is most likely due to surface desorption from the cracker device and its
connection tubing to the valves. This can be derived from the ‘He blank’, which is only halve
the size of the ‘cracker blank’ since the tube cracker had not ‘seen’ CO2 for one hour, i.e. desorption from the surface declined. It is important to note that as the relative contribution from
the cracker-GC-IRMS system is relatively large (45% from the total), and due to the additive
character of the blanks, the smaller contributions from consecutive steps represent coarse es-
Table 3-6 Step by step contribution of blank CO2 during the analysis starting (separated for the cracker-GCIRMS system and the sublimation extraction apparatus). The first column lists five blank experiments described above in sequence of increasing components involved during the analysis (with number of measurements in brackets). Second column: absolute amount of CO2 from each experiment. As the blanks are expected
to be cumulative during the analysis, the right column shows the relative fraction of each blank experiment.
number of blank experiment
amount of CO2
[nmol]
relative contribution
to total blank [%]
cracker-GC-IRMS only:
(1) ‘He-blank’ (n= 10)
0.004±0.005
~15
(2) ‘cracker-blank’ (n= 20)
0.015±0.005
~30
(3) ‘empty tube blank’ (n= 10)
0.02±0.005
~15
(4) ‘CO2 trap blank’ (n= 5)
0.025±0.005
~20
(5) ‘sublimation blank’ (n= 10)
0.03±0.005
~20
+ sublimation extraction:
Methods and instruments for δ13C and CO2 analysis on ice cores
99
timates. The total sublimation blank amounts to roughly 0.03 nmol CO2 measured from the
sublimation of ~5 g ‘blank ice’. For comparison, the amount of CO2 extracted from a Holocene ice core sample is ~6 nmol (~5 g ice with 10% volumetric air content and ~280 ppmv
CO2), thus, the contribution of the blank is 1-2 ppmv. The blank numbers 3-5 measured for
the sublimation extraction do not necessarily imply that the amount of CO2 for ice core samples is increased by this blank. It is more likely that during the preconditioning step of the
sample preparation any ‘outside CO2’ adsorbed onto surfaces is replaced by sample CO2,
which then gradually desorbs and exchanges during the sublimation of an ice sample. Likewise, the impact of the measured CO2 blank on the δ13C values of samples and standards can
only be estimated since the small blank peaks are out of the range for which δ13C values can
be reliably calculated. For a first order estimation of the blank’s influence on the δ13C measurement using an isotopic mass balance calculation one can assume that the δ13C value of the
blank is range of ±5‰ of the sample value. The resulting δ13C shift from 0.03 nmol blank on
a typical sample would then account for ±0.04‰.
4 Results and discussion of ice core measurements
In the following Chapter first results from ice core measurements using the newly developed
analysis technique will be presented. The key objective was to clarify, whether the measured
small-scale δ13C fluctuations, i.e. differences on a few centimeters or smaller than a year, reported by Eyer (2004) are a real feature or not. In the study by Eyer (2004) the average standard deviation of the measurement for the Holocene ice of the EDML ice core was 0.23‰.
Surprisingly, results on the EDC core had a scatter that was only half the value of the EDML
core. Furthermore, both δ13C records deviated also with respect to their mean values, with
EDML being ~0.2‰ more negative. Neither these high δ13C fluctuations on a sub-annual
scale, nor an absolute off-set between two ice cores on the same hemisphere are plausible for
a well mixed atmospheric parameter with a life time of decades or longer. Consequently, Eyer
(2004) speculated that an up to now not identified fractionation process during the bubble
close-off might be responsible for this phenomenon. For other gas species, evidence from firn
air measurements emerged, pointing to a size dependent fractionation (Huber et al., 2006;
Severinghaus and Battle, 2006). Yet, no indication was found that CO2 or the isotopic composition of CO2 might be affected by this physical fractionation process as well. Nevertheless,
the disappointing findings from Eyer (2004) somehow questioned the integrity of ice core
δ13C information as representative for the paleoatmosphere.
This Chapter is organized in the following way: Chapter 4.1 provides a brief description of
the measured ice core samples, followed by an overview of the main characteristics of the
measured δ13C and CO2 concentration data showing the results of one sample depth (Chapter
4.2). Next, the following three main issues will be discussed. First, the performance of the
newly developed analysis technique is evaluated on basis of real ice core samples and the results are compared with the performance of the air standard measurements (Chapter 4.3). Secondly, the ice core data are analyzed to address the question of the small-scale δ13C variability. Here, the differences among the sample replicates are considered (Chapter 4.4). Thirdly,
the absolute values for δ13C and the CO2 concentration are compared with existing records to
assess the accuracy of this analysis technique (Chapter 4.5).
4.1 Ice core samples from the EDML ice core
Based on the findings of Eyer (2004), the first crucial application of this newly developed
method was to shed light on this small-scale δ13C scatter. To this end, sample replicates were
analyzed on the EDML ice core, the core for which Eyer (2004) found the largest δ13C scatter.
Four depth intervals were selected for this study: 151 m, 253 m, 420 m, and 1056 m. While
the first three depths are of Holocene age and the ice quality is pure bubble ice, the samples
101
Results and discussions of ice core measurements
from the 1056 m depth interval correspond to the glacial period (Fig. 4-1). At this depth bubbles disappeared and most of the air content has already transformed to clathrates.
Figure 4-1 gives an impression of the time interval a single ice sample of 4.2 cm comprises in
terms of age. For the Holocene samples, between 0.5 and 0.8 years of ice are covered within a
single sample, for the glacial ice, the value is 1.5 years. However, these values correspond to a
time interval for the ice matrix, thus, to the time scale of dissolved and particulate impurities
in the ice. For the enclosed air the age difference between adjoining samples is virtually zero
due to the bubble enclosure processes during firnification (see Chapter 2.1). As the age distribution at the bubble close-off depth at EDML is ~59 years (see Fig. 2-3), ice samples that are
only centimeters to a few decimeters apart effectively contain identical air samples. From
each depth interval, three, four in case of depth 253 m, vertically adjoining ice core samples
were analyzed with the new method. As described above, each sample yields five sub-samples
in separate glass tubes. To achieve optimal measurement conditions to detect differences
among the three adjoining samples, all tubes of a depth interval were analyzed jointly within
one IRMS measurement session. Applying this design, potential differences among individual
40
2.5
35
2
30
25
1.5
20
1
15
Ice age [kyr BP]
Years within analysed ice core sample [yr]
IRMS runs, potentially introducing additional variance, were excluded.
10
0.5
5
0
0
200
400
600
800
Ice core depth [m]
1000
1200
0
Figure 4-1 Basic stratigraphic properties of the EDML ice core and its relation to the sample size to assess the
‘small-scale heterogeneity’. Left axis: number of years covered within the measured ice core sample of 4.2 cm
as a function of ice core depth. Right axis: Age of the ice as a function of depth (U. Ruth, personal communication, 2006). The general trend of the blue curve towards more years covered within a single ice sample is due to
the thinning rate with increasing depth and elapsed time. Superimposed on this trend is the shift in accumulation rate due to climate changes.
102
Results and discussions of ice core measurements
4.2 Main characteristics of the measured ice core data
Before the results of the measured ice samples are analyzed in detail, the data set of one depth
interval (420 m) is shown to visualize the general measurement features (Fig. 4-2). Shown are
the combined results for the δ13C values (upper panel) and the CO2 concentrations (lower
panel) for three vertical replicates (sample 1, 2, 3). Three general features are readily visible:
(1) Systematic differences among the three samples replicates are not observed. (2) Fairly
good reproducibility within a tube number class. (3) Systematic differences among the five
tube number classes are especially pronounced for δ13C values.
δ13C [‰]
−6
−6.5
sample 1 μ (σ): −6.50 (0.14)
sample 2 μ (σ): −6.44 (0.21)
sample 3 μ (σ): −6.47 (0.13)
weighted mean of samples
−7
−7.5
1
2
3
Tube No.
4
5
280
2
CO [ppmv]
275
270
sample 1 μ (σ): 275.2 (1.0)
sample 2 μ (σ): 274.8 (2.9)
sample 3 μ (σ): 274.5 (1.6)
weighted mean of samples
265
260
255
1
2
3
Tube No.
4
5
Figure 4-2 Example showing the measurement results of three ice core replicates from the depth interval ‘420
m’. Upper panel: δ13C values grouped according to the tube number, i.e. the sequence of sub-samples collected
during the sublimation of an ice sample. Visible is the marked deviation of the first tube number from the
weighted mean 20 of all sample tubes (dotted line), while tube number 2, 3, and 4 have almost the same level.
As a general feature, no systematic differences were observed among the three samples (weighted mean and
standard deviation of tube 1-5 are given in the legend). Where means deviated among the three samples it was
due to erratic ‘outliers’, which were not removed from the data set when no obvious reason during the analysis
procedure could be identified. Bottom panel: In this example CO2 concentration results show little dependence
on the tube number, but a somewhat increased scatter for tube number 1.
20
for averaging δ13C values and CO2 concentrations weighted means were calculated further on, since the corresponding air volume to which these values refer is not equal for all tubes (see Fig. 3-16 for details)
103
Results and discussions of ice core measurements
4.3 Systematic differences during the sublimation extraction
This paragraph focuses on potential processes to explain the observed differences among the
five tube numbers. As shown above for the sample depth 420 m (Fig. 4-2), the δ13C values of
tube number 1 are systematically lighter compared to the other tubes. This observation is valid
for all measurements. On average, the δ13C values for tube number 1 are 0.25‰ isotopically
lighter than the overall δ13C mean as summarized in Table 4-1. Moreover, slightly lighter
δ13C values were found as well for tube number 5. Additionally, the CO2 concentrations for
tube number 5 are on average 3.5 ppmv lower than the mean over all five tube numbers (see
Tab. 4-2). As the marked δ13C anomaly for tube number 1 is not observable for air standard
Table 4-1 Compilation of the δ13C results for the ‘vertical replicates’ of the four measured depths intervals.
From each depth interval three vertically adjoining ice core samples (replicates) were analyzed and corrected as
described in Chapter 3.8 except for the gravitation effect, which is not relevant at this stage. The standard deviation (1σ) and average for three replicates were calculated for each tube number class separately. To characterize the systematic difference among the five tube numbers, Δaverage, the deviation of each tube number mean
from the total average was calculated. The standard deviations of the five tube number classes for each depth
interval were averaged to characterize the scatter between the three sample replicates. A total average was
calculated for the standard deviations of each tube number to characterize the reproducibility for each tube
number. Furthermore, for each depth interval an average is calculated for the standard deviations and means
over tube numbers 1-5 (right column) to provide a measure for the scatter among the three replicate samples
and to allow a first comparison of the samples’ absolute values prior to the gravitation correction.
tube number (sub-sample during the sublimation)
ice core depth interval
151 m
standard deviation [‰]
mean [‰]
Δ average [‰]
253 m
standard deviation [‰]
mean [‰]
Δ average [‰]
420 m
standard deviation [‰]
mean [‰]
Δ average [‰]
1056 m
standard deviation [‰]
mean [‰]
Δ average [‰]
total averages
standard deviation [‰]
Δ average [‰]
1
2
3
4
5
0.05
-6.66
-0.32
0.07
-6.34
0.00
0.08
-6.15
0.19
0.11
-6.23
0.11
0.05
-6.32
0.02
0.07
-6.34
1.00
0.12
-6.55
-0.30
2.00
0.06
-6.21
0.04
3.00
0.07
-6.16
0.09
4.00
0.05
-6.20
0.05
5.00
0.04
-6.14
0.11
0.07
-6.25
1.00
0.03
-6.69
-0.22
2.00
0.03
-6.39
0.09
3.00
0.12
-6.31
0.17
4.00
0.07
-6.41
0.06
5.00
0.06
-6.56
-0.09
0.06
-6.47
1.00
0.06
-6.46
-0.15
2.00
0.00
-6.23
0.08
3.00
0.01
-6.17
0.14
4.00
0.05
-6.27
0.04
5.00
0.09
-6.42
-0.11
0.04
-6.31
0.07
-0.25
0.04
0.05
0.07
0.15
0.07
0.07
0.06
-0.02
average
0.06
104
Results and discussions of ice core measurements
measurements, the cause must be either related to a property of the ice core sample itself. Or,
the anomaly is caused by a process unique to the conditions occurring during the sublimation
of an ice core sample; i.e. a process that is not properly simulated by the introduction of air
standard to ‘blank ice’ with the reference gas inlet. However, a deviation for tube number 1 to
slightly more negative δ13C values was also detected for air standards, but this difference was
well within the envelope of the 1 σ measurement reproducibility (Fig. 3-17).
The option that the systematic δ13C differences among the tube numbers do reflect a primary
signal within the ice itself, like the postulated small-scale δ13C variability (Eyer, 2004), is
unlikely. For a layered ice core, the preferred concentration gradients for chemical compounds as well as changes in physical properties (e.g. the total air content or crystal size) are
Table 4-2 Compilation for the CO2 concentrations of the four measured depths intervals. The standard deviation
(1σ) and average for the three replicates were calculated for each tube number class separately. To characterize
the systematic difference among the five tube numbers, Δaverage, the deviation of each tube number mean from the
total average was calculated. The standard deviations of the five tube number classes for each depth interval
were averaged to characterize the scatter between the three ice core replicates. A total average was calculated for
the standard deviations of each tube number to characterize the reproducibility for each tube number. Furthermore, for each depth interval an average is calculated for the standard deviations and means over tube numbers
1-5 (right column) to provide a measure for the scatter between the three replicate samples, which allows a first
comparison of the samples’ absolute values prior to gravitation correction.
tube number (sub-sample during the sublimation)
ice core depth interval
1
2
3
4
5
1.0
2.0
3.0
4.0
5.0
standard deviation [ppmv]
mean [ppmv]
Δ average [ppmv]
3.0
286.1
4.3
281.2
2.1
280.8
5.1
281.4
2.8
277.0
4.8
-0.1
-0.5
0.1
-4.3
253 m
1.0
3.4
278.4
1.7
2.0
3.4
277.7
1.0
3.0
3.9
277.4
0.8
4.0
2.2
278.1
1.5
5.0
5.1
271.7
-5.0
3.6
276.7
1.0
3.5
273.6
-1.2
2.0
1.3
276.2
1.4
3.0
0.5
275.4
0.6
4.0
2.1
274.5
-0.3
5.0
0.6
274.2
-0.5
1.6
274.8
1.0
0.6
192.7
-4.1
2.0
2.4
196.3
-0.6
3.0
1.1
205.4
8.5
4.0
1.5
197.4
0.5
5.0
6.5
192.5
-4.3
2.4
196.9
2.6
0.3
2.9
0.4
1.9
2.4
2.7
0.5
3.7
-3.5
151 m
standard deviation [ppmv]
mean [ppmv]
Δ average [ppmv]
420 m
standard deviation [ppmv]
mean [ppmv]
Δ average [ppmv]
1056 m
standard deviation [ppmv]
mean [ppmv]
Δ average [ppmv]
average
3.5
281.3
total averages
standard deviation [ppmv]
Δ average [ppmv]
2.8
Results and discussions of ice core measurements
105
perpendicular to the horizontal plane. Because the ice core samples retain their natural orientation during the sublimation (see Fig. 3-3), the sublimation proceeds almost circular and orthogonal to the layering. Thus, for any time step equal amounts of ice are removed from each
horizontal layer. Therefore, an ordinary ice signal cannot explain the observed δ13C deviations
among the tubes.
More tricky are the next two scenarios, which assess whether the observed inter-tube δ13C
differences might reflect an apparent δ13C signal from the ice core sample. In this scenario,
the measured inter-tube δ13C differences are not regarded to reflect a primary, or ‘natural’
signal, but rather a post coring storage effect, or the result of the sample preparation. The fact
that the enclosed air in ice cores might partially be lost during the storage of ice cores was
first noticed during the analysis of the N2/O2 ratio (Bender et al., 1995). In case of the CO2
concentration, it is generally assumed that this storage effect is minor, and by removing a few
mm of ice prior to the analysis this problem is solved. Only recently Siegenthaler (2006) revisited this issue and modeled the effect of gas loss during the storage on the CO2 concentration. To this end, Siegenthaler (2006) used revised permeation coefficients of gas species
through ice, which are much higher than previously calculated (Ikeda-Fukazawa et al., 2004).
As shown in Figure 4-3, Siegenthaler (2006) found that CO2 becomes increasingly enriched
in the outer rim due to a preferential loss of gas species with higher permeabilities (mainly O2
and N2). So far, there is no evidence that also an isotopic fractionation might be connected
Figure 4-3 Modeling results conducted by Siegenthaler (2006) to estimate the effect of selective gas loss during the storage of ice cores on its CO2 concentration. The two curves illustrate the relative CO2 increase for two
years for an ice cylinder with a diameter of 5 cm at -25 °C using two permeation coefficients (previous assumptions: solid line; new assumptions: dashed line; see details in Siegenthaler (2006)). Note that the calculated CO2
increase for the outer part is not due to an inward diffusion from the ambient air as the absolute CO2 pressure
within the ice core is considerably higher than the outside pressure.
106
Results and discussions of ice core measurements
with this effect, i.e. whether δ13C values might also be affected. Because for this study at least
4 mm of ice were removed from the outer part, this effect is unlikely to explain the observed
δ13C anomaly of the first tube number. Likewise, the CO2 concentrations for the first tube
number are not notably elevated, except for the depth interval 151 m (see Tab. 4-1). Therefore, this storage effect cannot explain the observed inter-tube differences. Nevertheless, the
gas extraction technique used here, which successively penetrates from the outer part into the
inner part of an ice sample, is certainly especially vulnerable for such an effect (see Fig. 3-3).
In contrast, with mechanical extraction techniques the entire ice sample is extracted at once
and then yields a single value (‘mean of the ice piece’). In this case, the amount of air extracted from a possibly fractionated outer part will then be diluted by the bulk, thus, masking
this effect.
The second scenario assumes that during the sample preparation (vacuum pumping for two
hours) a selective gas loss occurs from the ice sample’s outer sheath. Craig et al. (1988) observed a gas loss induced by pumping at vacuum conditions during the analysis of fractured
ice samples with many visible micro cracks. As this fractionation effect involves molecular
diffusion, besides an elemental fractionation, also an isotopic fractionation will occur. As a
consequence, this causes an isotopic enrichment of the remaining gas while the escaped gas,
in turn, becomes isotopically depleted. In fact, Craig et al. (1988) found enriched δ13C(CH4)
values for fractured ice core samples and also Schaefer (2005) reported a higher scatter for
δ13C(CH4) values associated with fractured ice core samples. However, the deviation for tube
number 1 is in the opposite direction to this scenario, and the samples were inspected for visible cracks prior to the sample preparation and none of the analyzed samples had visible
cracks.
However, such cracks might form during the sublimation process when the ice is irradiated
with IR light, a phenomenon reported also by others (Güllük et al., 1998). Depending on the
light’s wavelength, the IR energy penetrates several mm to cm into the ice sample until it is
absorbed, whereby increasing the temperature. In contrast, energy can be removed from the
ice sample only at the sublimating surface by the latent heat flux. After a while the incoming
IR flux is balanced by the latent heat flux until a steady state is reached. In consequence, one
can imagine that within the ice sample a temperature gradient will establish due to differential
heat supply and removal. In accordance with this, at the method’s early development stage,
ice samples sometimes broke suddenly apart in several pieces during the ongoing sublimation.
The disintegration of ice samples during the sublimation was also described by Güllük et al.
(1998) as a common phenomenon at the beginning of the sublimation procedure when the ice
sample had an inhomogeneous temperature distribution. However, for the applied sublimation
settings used in this work, a sample break-up did not occur anymore, as the energy supply is
smoothly increased during the sublimation (see Tab. 3-3 in Chapter 3-3). Yet, it is not clear,
Results and discussions of ice core measurements
107
whether micro cracks might form within the IR absorbing outer part of the ice sample and if,
whether this can induce a measurable effect on the released gas composition given the short
time scales. Though possible effects from the latter scenario cannot be easily ruled out, the
fact that the analytical reproducibility for tube number 1 is as good as for the other tubes does
not concur with a cracking process (see Tab. 4-1 and 4-2). As a process like cracking of an ice
sample should be erratic, one would assume that the measured data are more randomly scattered, i.e. should lead to a higher standard deviation.
While the above discussed processes focused on the ice sample, sorption processes within the
sublimation apparatus will be examined in the following:
First the desorption of CO2 from the surfaces of the apparatus is considered. CO2
might desorb from surfaces at the beginning of the sublimation either due to a higher H2O
partial pressure by replacing CO2 from adsorption sites. Or, due to a temperature increase of
the glass vessel caused by the IR irradiation. Both effects were observed in previous studies
and ascribed to measured CO2 concentration off-sets (Zumbrunn et al., 1982; Güllük et al.,
1998; Siegenthaler, 2002). When desorption or outgassing takes place, the CO2 concentration
would increase and, in turn, δ13C is expected to decrease if one assumes desorption of ‘outside’ CO2. Ambient air within the laboratory is usually isotopically lighter than the sample by
about several permil. If this scenario were plausible, the δ13C shift of on average 0.25‰ observed for the first tube number should be in line with an isotopic mass balance calculation.
However, only the two uppermost sample depths (151 m and 253 m) show elevated CO2 concentrations for the first tube number at all (see Δaverage values in Tab. 4-2), while in the other
cases the CO2 concentrations are lower than the overall average. Applying an isotopic mass
balance for the 151 m depth interval, the 4.8 ppmv positive off-set connected with a negative
δ13C off-set of 0.31‰ would be in line with a δ13C value of the added CO2 of -25‰. For the
depth interval 253 m, with Δaverage values of 1.7 ppmv and 0.30‰ (see Tab. 4-2), the δ13C
value of the added CO2 would be -50‰. Such negative δ13C values are unrealistic for the current atmospheric δ13C level of -8.5‰ and considering the air exchange rate within rooms.
Secondly, the adsorption of CO2 onto the surfaces of the apparatus is discussed. As
CO2 is preferentially lost by this process, the CO2 concentration will decrease. If this adsorption process is connected either with a kinetic or an equilibrium fractionation, then also the
δ13C value of the remaining fraction becomes affected. Unfortunately, fractionation factors for
carbon isotopes for adsorption processes of CO2 are not available from the literature. Similar
to the isotopic mass balance calculation from above, huge and unrealistic fractionation factors
must be assumed to explain an 0.3‰ shift for a CO2 loss of only a few ppmv.
Therefore, neither assuming desorption from surfaces nor adsorption onto surfaces can conclusively explain the observed δ13C anomaly of tube number 1. The same is true for the less
pronounced deviation observed for tube number 5. Likewise, δ13C values are slightly more
108
Results and discussions of ice core measurements
negative and CO2 concentrations lower for the last tube number (Tab. 4-1 and 4-2). As described for air standard measurements in Chapter 3.8, small nonlinearities during the mass
spectrometric detection and the consecutive calculation steps might affect the δ13C and CO2
concentration results. Further analytic efforts are needed to identify the physical reasons for
the observed systematic differences among the five tube numbers. As a pragmatic step, both
the first and the last tube number were excluded from the calculations for the absolute values.
The three tubes collected in the middle of the continuous sublimation, numbers 2, 3 and 4,
reproduce well. Therefore, the averages of these central sub-samples are used for the comparison with other ice core data (Chapter 4-5). Note that the chance to identify such a bias during
the gas extraction process and afterwards to select the most trustworthy sub-samples is actually the strength of the approach developed in this work. Previous studies were dependent on
one single value being representative for the entire procedure. Thus, critical side effects at the
beginning or at the end of a procedure cannot be detected.
4.4 Small-scale variability – differences among replicates
One key question to be addressed by this first measurement campaign with three adjoining
sample replicates from four depth intervals is the issue of a small-scale δ13C variability. The
basic idea behind is, whether δ13C shows a variability on a temporal scale that is not of atmospheric origin. In other words, are there additional processes on the scale of centimeters
within the ice which add variability to the atmospheric δ13C signal? As described in Chapter
2, three physical processes are known to alter the atmosphere’s original δ13C value (gravitative settling, thermal diffusion, and diffusion along a concentration gradient). Though these
processes affect the δ13C values within the firn column and in the ice cores, none of these
processes is able to produce a δ13C variability on the centimeter scale.
The following calculation method is used to analyze the measured samples for a δ13C variability on a few centimeters. This approach is equivalent to the calculation of the method’s analytical reproducibility. However in its strict sense, the concept of analytical reproducibility
cannot be applied for this ice core analysis since the homogeneity of the sample material with
respect to δ13C is a priori not known. First, for each tube number class the standard deviation
(1σ) from the three sample replicates are calculated. This separate calculation for each tube
number class is essential due to the systematic differences observed among the different tube
numbers. This calculation is repeated for the four depth intervals. Secondly, the averaged
standard deviation for each of the five tube number classes is calculated. These total averages
range from 0.04‰ to 0.07‰, with an overall mean of 0.06‰ (see total averages at the bottom
of Tab. 4-1). Similarly, an averaged standard deviation is calculated for each depth interval
from the individual standard deviation of the five tube numbers. Here, the average standard
Results and discussions of ice core measurements
109
deviations range from 0.04 to 0.07‰, with an overall average of 0.06‰ (see averages in the
right column of Tab 4-1). Note that all five tube numbers were used for these calculations as
there is no evidence that the variability of tube number 1 and 5 is significantly higher compared to tube numbers 2, 3, and 4. As pointed out above, the reason for the observed δ13C offset for the first and the last tube number has not been identified. But irrespective of the nature
of this influence, it is save to assume that the variability of the replicates would rather be increased than decreased by adding an additional effect.
In effect, the average δ13C variability of 0.06‰ among three adjoining ice samples is not
higher than the reproducibility of air standard measurements as shown in Chapter 3.9. In turn,
this also implies that the ice core samples did not add to the overall measurement error derived from air standard measurements. Consequently, one can hypothesize for the EDML ice
core that the true δ13C variance along a few centimeters might in fact be lower than the measured variability (‘analytical reproducibility’) of 0.06‰.
Based on these replicate measurements from four selected depth intervals, there is no evidence to assume a small-scale δ13C variability within the EDML core. Hence, the large δ13C
scatter of 0.23‰, reported by Eyer (2004) for EDML, cannot be confirmed with the analytical
approach used in this work. However, besides numerous technical differences between the
two methods one critical aspect is the sample size used for the analysis. Eyer (2004) measured
on ice cubes with the dimensions of 2.5 cm x 2.5 cm x 1. 5 cm, whereby the shortest side with
1.5 cm was in direction of the vertical axis of the ice core. Hence, the vertical resolution was
1.5 cm, while in this study a sample comprises 4.2 cm of the ice core depth. Eyer (2004) argued that the large δ13C scatter he found in the EDML record might be due to an ice-internal
δ13C variability of unknown origin, but somehow related to the annual layering of the ice
core. Due to the small sample size of only 1.5 cm this signal was then resolved (Eyer, 2004).
As the vertical depth increment for this work is almost three times that of Eyer (2004), it is
conceivable that the smaller variability found in this study might be an effect of the larger
sample size. If the assumed small scale δ13C variability is a cyclical signal, then integrating
over a larger depth increment might explain this observation. On the other hand, if it is a randomly distributed δ13C variability, then the scatter should be invariant to the analyzed sample
size. However, even the larger samples used in this study are shorter than the length of an
annual layer in the Holocene. Furthermore, Holocene and LGM samples, which average over
different time intervals, have the same reproducibility. From the experiences of this work, it is
more likely that the large scatter found for the small sample size used in the work of Eyer
(2004) is rather related to an unrecognized equilibration or ‘side effect’, similar to the anomaly measured for the first tube number in this study.
110
Results and discussions of ice core measurements
4.5 Data comparison of the absolute values
In this paragraph the measured data from the EDML ice core will be compared with existing
δ13C and CO2 concentration time series from other ice cores. To compare and align the measured data with other ice core data two steps are crucial: First, correction for gravitational settling to account for fractionation in the firn column and secondly, the transfer of the data from
the primary depth scale to the gas-age scale. Both steps require special attention as two different methods were applied in the literature for the gravitational settling, and further, precise
dating of ice cores, especially the dating of the gas-age, is usually an ongoing effort. Besides
this, note that in all previous studies the measured δ13C values had been additionally corrected
for the isobaric interference of N2O during the mass spectrometric measurement, the called
N2O correction. As in this study CO2 and N2O were gas chromatographically separated prior
to the measurement, this correction does not apply to the δ13C data of this study, hence, no
uncertainty is introduced from this additional correction.
4.5.1 Gravitational correction and ice core dating
The effect of gravitational settling within the firn column on both the isotopic composition
and on mixing ratios of gas species can be either modeled with a firnification model, or determined empirically by measuring δ15N on the ice core. Only recently, Landais et al. (2006)
showed that although firnification models work fine for the Holocene conditions, they fail for
glacial conditions at low accumulation sites of East Antarctica. For the EDML core the re-
Figure 4-4 Comparison of measured δ15N values from the EDML ice core (grey diamonds) with model results
(black line) using a steady state approach calculating means for the Early Holocene and the Last Glacial Maximum (reproduced after Landais et al., 2006).
Results and discussions of ice core measurements
111
ported mismatch between modeled and measured δ15N values is ~0.1‰ for glacial conditions,
and ~0.05‰ for the Holocene (Fig. 4-4). Except for rapid climate changes, δ15N values can be
readily used to correct δ13C values as gravitational settling depends only on the absolute mass
difference. If rapid climate changes produce a significant temperature gradient within the
whole firn column, additionally thermal diffusion has to be taken into account (see Chapter
2.1). As rapid climate changes are not encountered in Antarctic ice (Blunier et al., 1998;
EPICA-community-members, 2006), gravitational correction has been reliably performed using δ15N. Unfortunately, for the EDML core δ15N values are only available for the depth intervals 550 m to 1200 m (see Figure 4-4, Landais et al., 2006). For the Holocene δ13C data, a
constant value of 0.44‰, derived from the youngest δ15N data of the Early Holocene, was
added to the measured δ13C data. This approach is justified since temperature and accumulation rate were pretty stable throughout the Holocene. For the sample depth 1056 m, a value of
0.39‰ is accessible from Landais et al. (2006). The standard deviation of the δ15N data is
0.006‰, thus, the attributed error from the gravitational correction is negligible compared to
the overall error of the δ13C analysis. The gravitational correction for the CO2 concentration is
analogue to δ13C and scales with δ15N due to the same physical principle (Eqn. 2-2 in Chapter
2.1). Accordingly, Eqn. 3-5 was applied (Etheridge et al., 1996):
CO2 ( gravity) =
CO 2 − CO 2 ⋅
δ 15N ⋅ 15.2
1000
(3 − 5)
with CO2(gravity) the corrected concentration, CO2 the measured data, and the conversion
factor 15.2 resulting from the mass difference between CO2 and the average mass of air. As
the absolute value of this correction is only ~1.8 ppmv for Holocene CO2 concentrations and
~1.3 ppmv for the Glacial the resulting error is again small (~0.02 ppmv) compared to the
analytical error.
Only recently, the dating and synchronization of the upper part of the EDML core was completed (EPICA-community-members, 2006). The time scales for the gas-age and ice-age were
derived from a synchronization of characteristic fluctuations in the methane concentration
between layer-counted Greenland ice cores and the EDML core. The basis of this approach is
that a change in the methane concentration constitutes a global atmospheric signal, though its
source lies mainly in the Northern Hemisphere. Further, the time scale of the EDC core is
based on a glaciological flow model. Via peak matching the two EPICA ice cores were synchronized using either prominent volcanic markers, or high dust concentrations. The final
outcome is a common time scale for Greenland and Antarctic ice cores, which will in the future allow to compare gas and ice records on the same time scale. The results of the ice core
analyses of this work are reported on this new time scale. However, the records with which
112
Results and discussions of ice core measurements
they are compared in the following were reported on individual and preliminary time scales,
thus, small dating errors between the records are expected to occur.
4.5.2 Comparison with other Antarctic ice core records
First attempts to measure δ13C on ice cores date back more than 20 years (Friedli et al., 1984;
Friedli et al., 1986; Siegenthaler et al., 1988), though, these early data covered very restricted
time periods and were not considered for the comparison. A first highly resolved and precise
δ13C data set was measured on the Law Dome ice core, however, it covers only the last 1000
years (Francey et al., 1999). Due to the high accumulation rate of more than 1 m at this
coastal site, this core is very precisely dated. In contrast, the Taylor Dome record (Indermühle
et al., 1999; Smith et al., 1999) spans more than 25 ka, but it lacks the high temporal resolution of the Law Dome record. The most recent δ13C results were measured on the two EPICA
cores, EDC and EDML (Eyer, 2004). Though both the EDC and the EDML δ13C records possess a high temporal resolution throughout the Holocene, and in case of EDC also the late
glacial, these records are difficult to interpret due to their high scatter and internal inconsistency.
Concerning the CO2 concentration more records are available compared to δ13C, therefore, the
most suitable CO2 records in terms of temporal resolution and precision were selected for the
comparison with the results from this study on the EDML core. From the same core, a highly
resolved CO2 concentration record measured by Siegenthaler et al. (2005a) covers the last
1000 years and provides an overlap with the uppermost sample depth of this study. A well
resolved and precise CO2 record from the EDC core covering the last 22 ka is used for reference by combining the EDC Holocene record (Monnin et al., 2004) with data from the Last
Glacial Termination (Monnin et al., 2001). For the older part, a CO2 record from Taylor
Dome (Indermühle et al., 2000) is used. However, the age scale for the Taylor Dome core is
under debate, thus, the alignment of the Taylor Dome records with the other Antarctic ice core
records should be made with caution (Monnin et al., 2004).
To compare the results from this study with these records from previous studies covering contrasting time frames and different resolutions, two separate figures are discussed in the following: Figure 4-5 covers the Holocene data from this work, while Figure 4-6 includes also
the data from the late glacial period. Within the uncertainty range, the CO2 concentration
measurements from this study agree well with the CO2 results from previous studies (lower
panels in Fig. 4-5 and Fig. 4-6). This is an important precondition to trust in the general suitability of the sublimation extraction technique also for the δ13C analysis. Note that all CO2
concentration records shown in Figure 4-5 and Figure 4-6 were retrieved using mechanical
extraction devices, which allow for only 60-85% extraction efficiency (see Chapter 3-2).
113
Results and discussions of ice core measurements
Generally, low extraction efficiencies are associated with clathrate ice when analyzed with
mechanical extraction techniques and special care has to be taken during the extraction of this
ice (Siegenthaler, 2006).
Problems during the analysis of clathrate ice for δ13C and CO2 concentrations were reported
by Eyer (2004). Using a mechanical cracker device, he noticed surprisingly low CO2 concentrations for the EDC core already below a depth of 650 m, the onset of the clathrate formation.
This issue is shown in Figure 4-6 (filled blue diamonds, lower panel), where the CO2 concentrations for ages older than 26 ka BP deviate towards lower values. In parallel, the corresponding δ13C values (filled blue diamonds, upper panel, Fig. 4-6) showed a higher scatter
and Eyer (2004) reported a linear trend towards more negative δ13C values with increasing
depth, which he corrected from the δ13C data. The comparison of the Holocene δ13C data indicates that the values from this study might be shifted towards more negative δ13C values
compared to previous studies (Fig. 4-5). Given the large scatter of the previous studies and the
-6
-6.2
13
δ C [‰]
-6.4
-6.6
-6.8
-7
Law Dome
EDML (Eyer04)
EDC (Eyer04)
Taylor Dome
EDML (this study)
-7.2
-7.4
Law Dome
EDC
EDML (Siegenthaler05)
Taylor Dome
EDML (Eyer04)
EDC (Eyer04)
EDML (this study)
CO 2 [ppmv]
320
300
280
260
1
2
3
Age [ka BP]
4
5
6
Figure 4-5 Comparison of Holocene δ13C values and CO2 concentrations analyzed on the EDML ice core described in this study (red squares) with results from previous ice core studies. Top: δ13C data from Law Dome
(black circles, Francey et al., 1999), from Eyer (2004) for the EDML core (blue open diamonds) and the EDC
core (blue filled diamonds), Taylor Dome (green circles, Indermühle et al., 1999). Bottom: CO2 concentrations
from Law Dome (black circles, Francey et al., 1999), EDC (grey diamonds, Monnin et al., 2004), EDML (blue
open circles, Siegenthaler et al., 2005a), Taylor Dome (green circles, Indermühle et al., 1999), and the corresponding CO2 concentrations for the δ13C data from Eyer (2004) for the EDML core (blue open diamonds) and
EDC (blue filled diamonds).
114
Results and discussions of ice core measurements
limited number of δ13C values of this study, an assumed off-set certainly needs confirmation
by further measurements. Nevertheless, a comparison of the shallowest sample depth corresponding to around 1000 years BP, with the δ13C value from Law Dome at the same age, suggests an off-set of 0.25‰.
There is little doubt in the absolute accuracy of the δ13C values measured on the Law Dome
ice core due to an overlap and good agreement of the ice core results with measurements
made on firn air and archived atmospheric air samples (see details in Francey et al., 1999). If
-6
-6.2
-6.4
13
δ C [‰]
-6.6
-6.8
-7
-7.2
Law Dome
-7.4
EDML (Eyer04)
-7.6
EDC (Eyer04)
Taylor Dome
Byrd
EDML (this study)
Law Dome
310
EDC
290
Taylor Dome (1999)
Taylor Dome (2000)
CO2 [ppmv]
270
EDML (Eyer04)
250
EDC (Eyer04)
230
EDML (this study)
210
190
170
150
130
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
36
38
A ge [ka BP]
Figure 4-6 Comparison of δ13C values and CO2 concentrations from the EDML core analyzed in this study (red
squares) with previous studies reaching back to 40 ka BP. Top: δ13C data from Law Dome (black circles, Francey
et al., 1999), from Eyer (2004) for the EDML core (blue open diamonds) and EDC (blue filled diamonds), Taylor
Dome (green circles, Indermühle et al., 1999; Smith et al., 1999), and Byrd ice core (yellow circles, Leuenberger et
al., 1992). Note that for EDC data older than 26 ka BP Eyer (2004) used a δ13C correction to account for the lower
extraction efficiency of clathrate ice. Bottom: CO2 concentrations from Law Dome (black circles, Francey et al.,
1999), EDC (grey diamonds, Monnin et al., 2001; Monnin et al., 2004), ‘Taylor Dome (1999)’ are the corresponding CO2 values to the δ13C data in the upper panel (green circles, Indermühle et al., 1999; Smith et al., 1999), ‘Taylor Dome (2000)’ are additional CO2 concentration measurements on older sections of the Taylor Dome core (green
open circles, Indermühle et al., 2000) and the corresponding CO2 values for the δ13C data in the upper panel from
Eyer (2004) for the EDML core (blue open diamonds) and EDC (blue filled diamonds). Note that the EDC data
older than 26 ka BP were measured on ice with clathrate formation resulting in lower CO2 values (shown are the
uncorrected values, see Eyer (2004) for details).
Results and discussions of ice core measurements
115
all δ13C data of this work were shifted by around 0.25‰ towards heavier δ13C values, the
three data points within the Holocene would better fit into the envelope of the δ13C results
from previous studies (see Fig. 4-5). However by doing so, the δ13C data point at 22 ka BP
would then be isotopically too heavy and the δ13C value will be in conflict with basic assumptions concerning the global carbon cycle during the glacial. Besides the multitude of processes
involved with the carbon cycle (e.g. Köhler and Fischer, 2004), the common understanding of
the paleoclimate community assumes lighter δ13C values for the glacial atmosphere-ocean
reservoir compared to the Holocene. Estimations about the reservoir size of the terrestrial carbon stock showed that about 500 Pg less carbon were stored on the continents as organic carbon during the glacial period, thus, isotopically light carbon was transferred to the atmosphere-ocean system (e.g. Adams and Faure, 1998; Joos et al., 2004). These assumptions
match δ13C measurements on marine carbonates, which were found to be more isotopically
depleted during glacial periods compared to the Holocene (Hodell et al., 2003).
As the available δ13C information from ice cores for the glacial period is sparse and for ages
older than ~20 ka BP highly noisy due to analytical limitations, there is no firm data basis to
judge whether the glacial value from his work is plausible or not (upper panel Fig. 4-6). For
the Holocene measurements, however, the observed δ13C discrepancy with the Law Dome
record calls for a critical evaluation of a potential off-set due to methodological differences
between these two approaches. These differences fall into two main categories: (1) due to the
difference in the extraction technique – mechanical vs. sublimation, (2) due the applied referencing procedure.
(1) Besides various technical details in the entire analysis, above all, it is the used extraction
technique that makes the largest difference between this study and all previous δ13C ice core
studies. Sublimation under vacuum not only releases the air trapped in bubbles or enclosed in
clathrates, but also any additional fraction of CO2 located within the ice matrix. In contrast,
with mechanical techniques CO2 is only extractable when it is either located in air bubbles, or
when the clathrates decompose and release their gases. If the ice contains any additional CO2,
then this fraction should produce an off-set in the CO2 concentration between the two extraction techniques. Two possible candidates for this ‘extra CO2’ were discussed in the past: First,
CO2 produced from chemical in-situ reactions (see Chapter 2.2. for details). Secondly, CO2
located within so called microbubbles, which, amongst other processes, might be produced
during the condensation of snow crystals (Siegenthaler et al., 2005a). While the δ13C values
for the chemically produced in-situ CO2 might strongly vary as discussed in Chapter 2.2, CO2
enclosed in these microbubbles should bear a δ13C value similar to the atmospheric δ13C signal. However, within the range of the measurement uncertainty (usually 2-3 ppmv), the CO2
concentrations measured with the sublimation technique do not deviate from the results obtained with mechanical devices (Siegenthaler et al., 2005b). Therefore, the amount of CO2
116
Results and discussions of ice core measurements
added by any of these two processes cannot exceed a few ppmv. Based on isotopic mass balance calculations (see Fig. 2-8 for details), such a small fraction of added carbon can barely
explain a δ13C shift of 0.25‰ assuming reasonable δ13C values for the added CO2. A further
argument against a possible in-situ contamination being the cause for the measured δ13C offset is the attained precision of the δ13C measurement on adjoining replicate samples (see
Chapter 4.4). As the chemical species in the EDML ice core show marked annual cycles and
the sample replicates cover less than an annual layer, an almost constant in-situ production of
CO2 is very unlikely to occur. From highly resolved CO2 concentration measurements on
Greenland ice cores, which are proven to be contaminated by in-situ CO2, it is known that this
in-situ CO2 co-varies with the precursor species on a seasonal time scale (Barnola, 1999;
Tschumi and Stauffer, 2000).
(2) As pointed out already, there is no doubt in the overall accuracy of the referencing strategy
of the Law Dome measurements due to the convincing overlap with atmospheric air measurements analyzed with the same analytical set-up. Unfortunately, this elegant approach is not
feasible with the continuous air release and trapping procedure applied in this work. Additionally, the isotopic composition of the air standard used in this study as reference basis for
the ice core results is certainly not optimal because the value of -2.7‰ deviates from the mean
value of the ice core samples (-6.7‰). An effect attributed to a difference in the δ13C value of
the air standard to the samples’ value is called ‘scale compression’. During intercomparison
studies of isotopic reference materials it was observed that laboratories using a working standard which deviated grossly from the sample are more likely to show a systematic off-set than
others (Verkouteren, 1999). This scale compression either results from actual calibration problems of the mass spectrometric measurement, or from an undetected contamination or sorption problem. In any case, the larger the δ13C difference for sample and standard, the larger is
the scale compression. For the ice core measurements planned in the near future, new air
standards, calibrated and certified by NOAA/CMDL 21, will be available to improve on the
current referencing situation. These two air standards have δ13C values of -7.9 and -8.4‰,
thus, only slightly isotopically lighter than typical samples. In turn, this enables to investigate
a potential scale compression effect since air standards of contrasting isotopic composition
can then be processed and compared. However, the general matrix difference between an ice
sample and the reference standard procedure, being a gas mixture admitted to a cube of pure
ice, still remains. To improve on this issue for future ice core measurements, a ‘reference ice
standard’ will be analyzed in conjunction with ice core samples to provide a second reference
basis additional to the air standards (see further details below).
21
Climate Monitoring and Diagnostics Laboratory (CMDL) of the National Oceanic and Atmospheric Administration (NOAA) at Boulder, Colorado
Results and discussions of ice core measurements
117
4.6 Outlook
The presented analysis technique for ice cores allowed for the first time to link a quantitative
extraction technique using sublimation under vacuum with a high-precision δ13C measurement. It could be shown that the previously postulated small scale δ13C heterogeneity found in
the EDML core, which challenged the δ13C approach, fortunately, could not be reproduced
with this work. The reproducibility of the measured ice core samples was on average 0.06‰,
which is considerably smaller than the average scatter of 0.23‰ for EDML ice, or 0.13‰ for
EDC ice reported by Eyer (2004). Having solved this key question, the next effort will clearly
focus on measuring a δ13C record throughout the Holocene to address the open questions
about the Holocene carbon cycle (Claussen et al., 2005; Ruddiman, 2005; Broecker and
Stocker, 2006). The argument is, whether the Holocene CO2 evolution with its minimum of
260 ppmv at 7 ka BP and its later rise to a preindustrial value of 280 ppmv was due to natural
equilibration of the ocean’s carbonate system as proposed by Broecker and Stocker (2006), or
due to early human activities, as proposed by Ruddiman (2005). In the latter case, the CO2
increase would be connected with a decrease in atmospheric δ13C due a net CO2 flux from
depleted terrestrial biomass into the atmosphere-ocean system. To answer this question, an
optimized measurement scheme will be necessary to resolve the expectedly small δ13C differences among the Holocene samples. With the measurement scheme applied above, which was
optimized to address the question of the small scale δ13C scatter, several replicates of one
sample depth were analyzed collectively within a single IRMS session. Though optimized to
resolve differences within a sample depth, the precision to resolve differences among different
sample depths was not optimal for this purpose. This results from the fact that for each session an individual off-set from air standard measurement is calculated and subtracted from the
ice core samples. Due to the uncertainty involved with this off-set (±0.06‰) and the risk of an
undetected bias among different measurement sessions, best results can be obtained when
different sample depths are measured within a single session. Further, a ‘reference ice standard’ should accompany each measurement session in addition to air standards to permit a
direct comparison between sample ice and ‘reference ice’ processed under truly identical conditions. Using an ice core standard as an additional reference basis is now feasible since (1)
this work demonstrated that the EDML ice carries a homogenous δ13C composition on the
centimeter-scale, thus, adjoining samples can be regarded to yield identical δ13C values; (2) a
newly drilled shallow ice core close to the EDML drill site allows to set aside a sufficient
quantity of young and dated ice for this reference purpose. Due to the gas-age overlap with
the δ13C data set from Law Dome (Francey et al., 1999), being the most trustworthy δ13C data
available so far, the measured Holocene δ13C data can then be additionally tied to the Law
Dome record. Defining a reference ice besides an air standard as a second tie point for δ13C
118
Results and discussions of ice core measurements
offers also the advantage that subtle δ13C drifts of the air standard within the pressure cylinder
can be identified and a change in the air standard can be better bridged. Further, it offers the
possibility for intercomparison studies among different laboratories and methods.
While measuring a precise δ13C record in the bubble ice of the Holocene can reveal the riddle
of the unresolved CO2 anomaly of the Early Holocene, the actual strength of this new method
lies in the climate history enclosed in clathrate ice. Only a small proportion of the atmosphere’s climate history is stored in bubble ice, thus is accessible with current mechanical extraction devices. For the EDML ice core bubble ice covers only the first ~700 m core section
or 15 ka from an age of around 150 ka in 2400 m (EPICA-community-members, 2006). The
grand questions to be answered with δ13C data are the identification and quantification of the
underlying processes driving the prominent glacial/interglacial CO2 difference of about 80
ppmv. Further, the EDC ice core reaches back to eight glacial cycles and offers the chance to
compare the driving processes between the large 100 ka cycles and the proceeding 40 ka cycles with so called lukewarm interglacials.
References
Adachi Y, Kawamura K, Armi L, Keeling RF. Diffusive separation of the lower atmosphere. Science 2006; 311:
1429.
Adams JM, Faure H. A new estimate of changing carbon storage on land since the last glacial maximum, based
on global land ecosystem reconstruction. Global and Planetary Change 1998; 16-17: 3.
Allison CE, Francey RJ, Meijer HAJ. Recommendations for the reporting of stable isotope measurements of
carbon and oxygen in CO2 gas. Reference and intercomparison materials for stable isotopes in light
elements. In: Proceedings of IAEA consultants' meeting, Vienna, Austria, 1-3 December 1993. IAEATECDOC-825, International Atomic Energy Agency, Vienna. 1995; 155.
Anklin M, Barnola J-M, Schwander J, Stauffer B, Raynaud D. Processes affecting the CO2 concentrations measured in Greenland ice. Tellus 1995; 47: 461.
Archer D, Winguth A, Lea D, Mahowald N. What caused the glacial/interglacial atmospheric pCO2 cycles?
Reviews of Geophysics 2000; 38: 159.
Arens NC, Jahren AH, Amundson R. Can C3 plants faithfully record the carbon isotopic composition of atmospheric carbon dioxide? Paleobiology 2000; 26: 137.
Arrhenius. On the Influence of Carbonic Acid upon the Temperature of the Ground. Philosophical Magazine 41
1896; 237.
Barnes PRF, Wolff EW, Mallard DC, Mader HM. SEM Studies of the Morphology and Chemistry of Polar Ice.
Microscopy Research and Technique 2003; 62: 62.
Barnola JM. Status of the atmospheric CO2 reconstruction from ice core analyses. Tellus, Series B: Chemical
and Physical Meteorology 1999; 51: 151.
Bender M, Sowers T, Barnola J-M, Chappellaz J. Changes in the O2/N2 ratio of the atmosphere during recent
decades reflected in the composition of air in the firn at Vostok Station, Antarctica. Geophysical Research Letters 1994; 21: 189.
Bender M, Sowers T, Lipenkov V. On the concentrations of O2, N2, and Ar in trapped gases from ice cores.
Journal of Geophysical Research 1995; 100:
Bender ML. Orbital tuning chronology for Vostok climate record supported by trapped gas composition. Earth
and Planetary Science Letters 2002; 204: 275.
Bernard S, Röckmann T, Kaiser J, Barnola JM, Fischer H, Blunier T, Chappellaz J. Constraints on N2O budget
changes since pre-industrial time from new firn air and ice core isotope measurements. Atmospheric
Chemistry and Physics 2006; 6: 493.
Bintanja R, Van De Wal RSW, Oerlemans J. Modelled atmospheric temperatures and global sea levels over the
past million years. Nature 2005; 437: 125.
Blunier T, Chappellaz J, Schwander J, Dällenbach A, Stauffer B, Stocker TF, Raynaud D, Jouzel J, Clausen HB,
Hammer CU, Johnson SJ. Asynchrony of Antarctic and Greenland climate change during the last glacial period. Nature 1998; 394: 739.
Brand WA. High precision isotope ratio monitoring techniques in mass spectrometry. Journal of Mass Spectrometry 1996; 31: 225.
Brenninkmeijer CAM. Automated Variable Temperature Liquid Nitrogen Cold Trap. Analytical Chemistry 1982;
54: 2622.
120
References
Brenninkmeijer CAM, Janssen C, Kaiser J, Röckmann T, Rhee TS, Assonov SS. Isotope Effects in the Chemistry of Atmospheric Trace Compounds. Chemical Reviews 2003; 103: 5125.
Broecker WS, Clark E. Holocene atmospheric CO2 increase as viewed from the seafloor. Global Biogeochemical
Cycles 2003; 17: 1052.
Broecker WS, Peng T-H. The role of CaCO3 compensation in the glacial to interglacial atmospheric CO2 change.
Global Biogeochemical Cycles 1987; 1: 15.
Broecker WS, Stocker TF. The Holocene CO2 Rise: Anthropogenic or Natural? EOS 2006; 87: 26.
Brovkin V, Hofmann M, Bendtsen J, Ganopolski A. Ocean biology could control atmospheric δ13C during glacial-interglacial cycle. Geochemistry, Geophysics, Geosystems 2002; 3: 1.
Brunet F, Gaiero D, Probst JL, Depetris PJ, Lafaye FG, Stille P. δ13C tracing of dissolved inorganic carbon
sources in Patagonian rivers (Argentina). Hydrological Processes 2005; 19: 3321.
Buchmann N, Brooks JR, Rapp KD, Ehleringer JR. Carbon isotope composition of C4 grasses is influenced by
light and water supply. Plant, Cell and Environment 1996; 19: 392.
Caillon N, Severinghaus JP, Barnola JM, Chappellaz J, Jouzel J, Parrenin F. Estimation of temperature change
and of gas age - Ice age difference, 108 kyr B.P., at Vostok, Antarctica. Journal of Geophysical Research D: Atmospheres 2001; 106: 31893.
Cerling TE, Quade J, Wang Y, Bowman JR. Carbon isotopes in soils and palaeosols as ecology and palaeoecology indicators. Nature 1989; 341: 138.
Chappellaz J, Blunier T, Kints S, Dällenbach A, Barnola J-M, Schwander J, Raynaud D, Stauffer B. Changes in
the atmospheric CH4 gradient between Greenland and Antarctica during the Holocene. Journal of Geophysical Research 1997; 102: 15987.
Chebbi A, Carlier P. Carboxylic acids in the troposphere, occurrence, sources, and sinks: A review. Atmospheric
Environment 1996; 30: 4233.
Ciais P, Denning AS, Tans PP, Berry JA, Randall DA, Collatz GJ, Sellers PJ, White JWC, Trolier M, Meijer
HAJ, Francey RJ, Monfray P, Heimann M. A three-dimensional synthesis study of δ18O in atmospheric
CO2 1. Surface fluxes. Journal of Geophysical Research D: Atmospheres 1997; 102: 5857.
Ciais P, Tans PP, White JWC, Trolier M, Francey RJ, Berry JA, Randall DR, Sellers PJ, Gollatz JG, Schimel
DS. Partitioning of ocean and land uptake of CO2 as inferred by δ13C measurements from the NOAA
Climate Monitoring and Diagnostics Laboratory Global Air Sampling Network. Journal of Geophysical
Research 1995; 100: 5051.
Claussen M, Brovkin V, Calov R, Ganopolski A, Kubatzki C. Did humankind prevent a Holocene glaciation?
Comment on Ruddiman's hypothesis of a Pre-Historic Anthropocene. Climatic Change 2005; 69: 409.
Colbeck SC. Air movement in snow due to windpumping. Journal of Glaciology 1989; 35: 209.
Coleman D. Fractionation in inlet system, December 9th 1996. Archives of [email protected],
http://list.uvm.edu/archives/isogeochem.html 1996;
Coleman DD. Tube Cracker for Opening Samples Sealed in Glass Tubing. Analytical Chemistry 1981; 53: 1962.
Craig H. Isotopic standards for carbon and oxygen and correction factors for mass-spectrometric analysis of
carbon dioxide. Geochimica et Cosmochimica Acta 1957; 12: 133.
Craig H, Horibe Y, Sowers T. Gravitational separation of gases and isotopes in polar ice caps. Science 1988;
242: 1675.
Curry WB, Duplessy JC, Labeyrie LD, Shackleton NJ. Changes in the distribution of δ13C of deep water ΣCO2
between the last glaciation and the Holocene. Paleoceanography 1988; 3: 317.
121
References
Dargaville RJ, Doney SC, Fung IY. Inter-annual variability in the interhemispheric atmospheric CO2 gradient:
Contributions from transport and the seasonal rectifier. Tellus, Series B: Chemical and Physical Meteorology 2003; 55: 711.
Delmas R. A natural artefact in Greenland ice-core CO2 measurements. Tellus 1993; 45: 391.
Des Marais DJ, Hayes JM. Tube Cracker for Opening Glass-Sealed Ampoules under Vacuum. Analytical Chemistry 1976; 48: 1651.
Ding ZL, Yang SL. C3/C4 vegetation evolution over the last 7.0 Myr in the Chinese Loess Plateau: Evidence
from pedogenic carbonate δ13C. Palaeogeography, Palaeoclimatology, Palaeoecology 2000; 160: 291.
EPICA-community-members. One-to-one interhemispheric coupling of millennial polar climate variability during the last glacial. Nature 2006; submitted.
Etheridge DM, Steele LP, Langenfelds RL, Francey RJ, Barnola J-M, Morgan VI. Natural and anthropogenic
changes in atmospheric CO2 over the last 1000 years from air in Antarctic ice and firn. Journal of Geophysical Research 1996; 101: 4115.
Eyer M. Highly resolved δ13C measurements on CO2 in air from Antarctic ice cores. Dissertation University
Bern 2004;
Ferretti DF, Lowe DC, Martin RJ, Brailsford GW. A new gas chromatograph-isotope ratio mass spectrometry
technique for high-precision, N2O-free analysis of δ13C and δ18O in atmospheric CO2 from small air
samples. Journal of the Geophysical Research 2000; 105: 6709.
Fischer H, Wahlen M, Smith HJ. Reconstruction of glacial/interglacial changes in the global carbon cycle from
CO2 and δ13CO2 in Antarctic ice cores. Memoirs of the National Institute for Polar Research 2003; 57:
121.
Fischer H, Wahlen M, Smith J, Mastroianni D, Deck B. Ice core records of atmospheric CO2 around the last
three glacial terminations. Science 1999; 283: 1712.
Flückiger J, Blunier T, Stauffer B, Chappellaz J, Spahni R, Kawamura K, Schwander J, Stocker TF, Dahl-Jensen
D. N2O and CH4 variations during the last glacial path epoch: Insight into global processes. Global Biogeochemical Cycles 2004; 18:
Flückiger J, Dällenbach A, Blunier T, Stauffer B, F. ST, Raynaud D, Barnola J-M. Variations in atmospheric
N2O concentration during abrupt climatic changes. Science 1999; 285: 227.
Francey RJ. Tasmanian tree rings belie suggested anthropogenic 13C/12C trends. Nature 1981; 290: 232.
Francey RJ, Allison CE, Etheridge DM, Trudinger CM, Enting IG, Leuenberger M, Langenfelds RL, Michel E,
Steele LP. A 1000-year high precision record of δ13C in atmospheric CO2. Tellus 1999; 51B: 170.
Francey RJ, Farquhar GD. An explanation of 13C/12C variations in tree rings. Nature 1982; 297: 28.
Freitag J, Dobrindt U, Kipfstuhl J. A new method for predicting transport properties of polar firn with respect to
gases on the pore-space scale. Annals of Glaciology 2002; 35: 538.
Freitag J, Wilhelms F, Kipfstuhl S. Microstructure-dependent densification of polar firn derived from X-ray
microtomography. Journal of Glaciology 2004; 50: 243.
Frey MM, Stewart RW, McConnell JR, Bales RC. Atmospheric hydroperoxides in West Antarctica: Links to
stratospheric ozone and atmospheric oxidation capacity. Journal of Geophysical Research D: Atmospheres 2005; 110: 1.
Friedli H, Lötscher H, Oeschger H, Siegenthaler U, Stauffer B. Ice core record of the
pheric CO2 in the past two centuries. Nature 1986; 324: 237.
13
C/12C ratio of atmos-
122
References
Friedli H, Moor E, Oeschger H, Siegenthaler U, Stauffer B. 13C/12C ratios in CO2 extracted from Antarctic ice.
Geophysical Research Letters 1984; 11: 1145.
Ghosh P, Brand WA. Stable isotope ratio mass spectrometry in global climate change research. International
Journal of Mass Spectrometry 2003; 228: 1.
Ghosh P, Brand WA. The effect of N2O on the isotopic composition of air-CO2 samples. Rapid Communications
In Mass Spectrometry 2004; 18: 1830.
Gillett RW, Van Ommen TD, Jackson AV, Ayers GP. Formaldehyde and peroxide concentrations in Law Dome
(Antarctica) firn and ice cores. Journal of Glaciology 2000; 46: 15.
Goldstein AH, Shaw SL. Isotopes of Volatile Organic Compounds: An Emerging Approach for Studying Atmospheric Budgets and Chemistry. Chemical Reviews 2003; 103: 5025.
Goujon C, Barnola JM, Ritz C. Modelling the densification of polar firn including heat diffusion: Application to
close-off characteristics and gas isotopic fractionation for Antarctica and Greenland sites. Journal of
Geophysical Research D: Atmospheres 2003; 108:
Grachev AM, Severinghaus JP. Laboratory determination of thermal diffusion constants for 29N2/28N2 in air at
temperatures from -60 to 0°C for reconstruction of magnitudes of abrupt climate changes using the ice
core fossil-air plaeothermometer. Geochimica et Cosmochimica Acta 2003; 67: 345.
Guimbaud C, Grannas AM, Shepson PB, Fuentes JD, Boudries H, Bottenheim JW, Domine F, Houdier S, Perrier
S, Biesenthal TB, Splawn BG. Snowpack processing of acetaldehyde and acetone in the Arctic atmospheric boundary layer. Atmospheric Environment 2002; 36: 2743.
Güllük T, Slemr F, Stauffer B. Simultaneous measurement of CO2, CH4, and N2O in air extracted by sublimation
from Antarctica ice cores: Confirmation of the data obtained using other extraction techniques. Journal
of the Geophysical Research 1998; 103: 15971.
Haan D, Raynaud D. Ice core record of CO variations during the last two millennia: atmospheric implications
and chemical interactions within the Greenland ice. Tellus, Series B: Chemical and Physical Meteorology 1998; 50B: 253.
Hall JA, Barth JAC, Kalin RM. Routine Analysis by High Precision Gas Chromatography/Mass Selective Detector/Isotope Ratio Mass Spectrometry to 0.1 Parts Per Mil. Rapid Communications In Mass Spectrometry
1999; 13: 1231.
Halsted RE, Nier AO. Gas Flow through the Mass Spectrometer Viscous Leak. The Review of Scientific Instruments 1950; 21: 1019.
Hobbie EA, Werner RA. Intramolecular, compound-specific, and bulk carbon isotope patterns in C3 and C4
plants: a review and synthesis. New Phytologist 2004; 161: 371.
Hodell DA, Venz KA, Charles CD, Ninnemann US. Pleistocene vertical carbon isotope and carbonate gradients
in the South Atlantic sector of the Southern Ocean. Geochemistry Geophysics Geosystems 2003; 4:
Honig RE. Gas Flow in the Mass Spectrometer. Journal of Applied Physics 1945; 16: 646.
Houdier S, Perrier S, Domine F, Cabanes A, Legagneux L, Grannas AM, Guimbaud C, Shepson PB, Boudries H,
Bottenheim JW. Acetaldehyde and acetone in the Arctic snowpack during the ALERT2000 campaign.
Snowpack composition, incorporation processes and atmospheric impact. Atmospheric Environment
2002; 36: 2609.
Huber C, Beyerle U, Leuenberger M, Schwander J, Kipfer R, Spahni R, Severinghaus JP, Weiler K. Evidence
for molecular size dependent gas fractionation in firn air derived from noble gases, oxygen, and nitrogen measurements. Earth and Planetary Science Letters 2006; 243: 61.
References
123
Hutterli MA, McConnell JR, Bales RC, Stewart RW. Sensitivity of hydrogen peroxide (H2O2) and formaldehyde
(HCHO) preservation in snow to changing environmental conditions: Implications for ice core records.
Journal of Geophysical Research D: Atmospheres 2003; 108:
Ikeda-Fukazawa T, Fukumizu K, Kawamura K, Aoki K, Nakagawa T, Hondoh T. Effects of molecular diffusion
on trapped gas composition in polar ice cores. Earth and Planetary Science Letters 2005; 229: 183.
Ikeda-Fukazawa T, Hondoh T, Fukumura T, Fukazawa H, Mae S. Variation in N2/O2 ratio of occluded air in
Dome Fuji Antarctic ice. Journal of Geophysical Research 2001; 106: 17799.
Ikeda-Fukazawa T, Kawamura K, Hondoh T. Mechanism of molecular diffusion in ice crystals. Molecular Simulation 2004; 30: 973.
Ikeda T, Fukazawa H, Mae S, Pepin L, Duval P, Champagnon B, Lipenkov VY, Hondoh T. Extreme fractionation of gases caused by formation of clathrate hydrates in Vostok Antarctic ice. Geophysical Research
Letters 1999; 26: 91.
Indermühle A, Monnin E, Stauffer B, Stocker TF, Wahlen M. Atmospheric CO2 concentration from 60 to 20 kyr
BP from the Taylor Dome ice core, Antarctica. Geophysical Research Letters 2000; 27: 735.
Indermühle A, Stocker TF, Joos F, Fischer H, Smith HJ, Wahlen M, Deck B, Mastroianni D, Tschumi J, Blunier
T, Meyer R, Stauffer B. Holocene carbon-cycle dynamics based on CO2 trapped in ice at Taylor Dome,
Antarctica. Nature 1999; 398: 121.
Joos F, Gerber S, Prentice IC, Otto-Bliesner BL, Valdes PJ. Transient simulations of Holocene atmospheric
carbon dioxide and terrestrial carbon since the Last Glacial Maximum. Global Biogeochemical Cycles
2004; 18:
Jouzel J, Vimeux F, Caillon N, Delaygue G, Hoffmann G, Masson-Delmotte V, Parrenin F. Magnitude of isotope/temperature scaling for interpretation of central Antarctic ice cores. Journal of Geophysical Research D: Atmospheres 2003; 108:
Kaspers KA, van de Wal RSW, van den Broeke MR, Schwander J, van Lipzig NPM, Brenninkmeijer CAM.
Model calculations of the age of firn air across the Antarctic continent. Atmospheric Chemistry and
Physics 2004; 4: 1365.
Kawamura K, Nakazawa T, Aoki S, Sugawara S, Fujii Y, Watanabe O. Atmospheric CO2 variations over the
last three glacial-interglacial climatic cycles deduced from the Dome Fuji deep ice core, Antarctica using a wet extraction technique. Tellus B 2003; 55: 126.
Kawamura K, Severinghaus JP, Ishidoya S, Sugawara S, Hashida G, Motoyama H, Fujii Y, Aoki S, Nakazawa
T. Convective mixing of air in firn at four polar sites. Earth and Planetary Science Letters 2006; 244:
672.
Kawamura K, Yokoyama K, Fujii Y, Watanabe O. A Greenland ice core record of low molecular weight dicarboxylic acids, ketocarboxylic acids, and dicarbonyls: A trend from little ice age to the present (1540 to
1989 A.D.). Journal of Geophysical Research D: Atmospheres 2001; 106: 1331.
Keeling CD, Whorf TP. Trends: A compendium of data on global change. In Carbon Dioxide Information
Analysis Center, Oak Ridge National Laboratory: Oak Ridge, TN, 2000;
Kennedy H, Kennedy DP. Simplified tube cracker for opening samples sealed in glass tubes while under vacuum. Analytical Proceedings Including Analytical Communications 1994; 31: 299.
Keppler F, Kalin RM, Harper DB, McRoberts WC, Hamilton JTG. Carbon isotope anomaly in the major plant
C1 pool and its global biogeochemical implications. Biogeosciences 2004; 1: 123.
Köhler P, Fischer H. Simulating changes in the terrestrial biosphere during the last glacial/interglacial transition.
Global and Planetary Change 2004; 43: 33.
124
References
Köhler P, Fischer H, Munhoven G, Zeebe RE. Quantitative interpretation of atmospheric carbon records over the
last glacial termination. Global Biogeochemical Cycles 2005a; 19:
Köhler P, Joos F, Gerber S, Knutti R. Simulated changes in vegetation distribution, land carbon storage, and
atmospheric CO2 in response to a collapse of the North Atlantic thermohaline circulation. Climate Dynamics 2005b; 25: 689.
Kuhs WF, Chazallon B, Radaelli PG, Pauer F. Cage occupancy and compressibility of deuterated N2-clathrate
hydrate by neutron diffraction. Journal of Inclusion Phenomena 1997; 29: 65.
Landais A, Barnola JM, Kawamura K, Caillon N, Delmotte M, Van Ommen T, Dreyfus G, Jouzel J, MassonDelmotte V, Minster B. Firn-air δ15N in modern polar sites and glacial-interglacial ice: a model-data
mismatch during glacial periods in Antarctica? Quaternary Science Reviews 2006; 25: 49.
Leckrone KJ, Hayes JM. Efficiency and Temperature Dependence of Water Removal by Membrane Dryers.
Analytical Chemistry 1997; 69: 911.
Leckrone KJ, Hayes JM. Water-Induced Errors in Continuous-Flow Carbon Isotope Ratio Mass Spectrometry.
Analytical Chemistry 1998; 70: 2737.
Leuenberger M, Siegenthaler U, Langway CC. Carbon isotope composition of atmospheric CO2 during the last
ice age from an Antarctic ice core. Nature 1992; 357: 488.
Leuenberger MC, Eyer M, Nyfeler P, Stauffer B, Stocker TF. High-resolution δ13C measurement on ancient air
extracted from less than 10 cm3 of ice. Tellus 2003; 55B: 138.
Lichtfouse E. Compound-specific isotope analysis. Application to archaeology, biomedical sciences, biosynthesis, environment, extraterrestrial chemistry, food science, forensic science, humic substances, microbiology, organic geochemistry, soil science and sport. Rapid Communications In Mass Spectrometry
2000; 14: 1337.
Linke H, Alemán BJ, Melling LD, Taormina MJ, Francis MJ, Dow-Hygelund CC, Narayanan V, Taylor RP,
Stout A. Self-propelled leidenfrost droplets. Physical Review Letters 2006; 96:
Lloyd J, Farquhar GD. 13C discrimination during CO2 assimilation by the terrestrial biosphere. Oecologia 1994;
99: 201.
Marino BD, McElroy MB, Salawitch RJ, Spaulding WG. Glacial-to-interglacial variations in the carbon isotopic
composition of atmospheric CO2. Nature 1992; 357: 461.
Mariotti A. Atmospheric nitrogen is a reliable standard for natural 15N abundance measurements. Nature 1983;
303: 685.
Meier-Augenstein W. Review: Applied gas chromatography coupled to isotope ratio mass spectrometry. Journal
of Chromatography A 1999; 842: 351.
Meier-Augenstein W, Watt P, Langhans C. Influence of Gas Chromatographic Parameters on Measurement of
13 12
C/ C Isotope Ratios by Gas-Liquid Chromatography-Combustion Isotope Ratio Mass spectrometry.
Journal of Chromatography A 1996; 752: 233.
Merritt DA, Hayes JM. Factors Controlling Precision and Accuracy in Isotope-Ratio-Monitoring Mass Spectrometry. Analytical Chemistry 1994; 66: 2336.
Monnin E. CO2 Konzentrationsmessungen an polaren Eiskernen insbesondere am Eisbohrkern Dome Concordia,
Antarktis. Master thesis, Physikalisches Institut, Universität Bern 2000;
Monnin E, Indermühle A, Dällenbach A, Flückiger J, Stauffer B, Stocker TF, Raynaud D, Barnola J-M. Atmospheric CO2 concentration over the last termination. Science 2001; 291: 112.
Monnin E, Steig EJ, Siegenthaler U, Kawamura K, Schwander J, Stauffer B, Stocker TF, Morse DL, Barnola
JM, Bellier B, Raynaud D, Fischer H. Evidence for substantial accumulation rate variability in Antarc-
References
125
tica during the Holocene, through synchronization of CO2 in the Taylor Dome, Dome C and DML ice
cores. Earth and Planetary Science Letters 2004; 224: 45.
Mook WG. 13C in atmospheric CO2. Netherlands Journal of Sea Research 1986; 20: 211.
Mook WG. Environmental Isotopes in the Hydrological Cycle; Principles and Applications (Volume I: Introduction - Theory, Methods, Review). UNESCO/IAEA, Paris, 2000;
Narukawa M, Kawamura K, Li SM, Bottenheim JW. Dicarboxylic acids in the Arctic aerosols and snowpacks
collected during ALERT 2000. Atmospheric Environment 2002; 36: 2491.
Norton GA. Improved tube cracker for opening vacuum-sealed glass tubes. Radiocarbon 2005; 47: 177.
O'Leary MH. Carbon isotope fractionation in plants. Phytochemistry 1981; 20: 553.
Ohno H, Igarashi M, Hondoh T. Salt inclusions in polar ice core: Location and chemical form of water-soluble
impurities. Earth and Planetary Science Letters 2005; 232: 171.
Ohno H, Igarashi M, Hondoh T. Characteristics of salt inclusions in polar ice from Dome Fuji, East Antarctica.
Geophysical Research Letters 2006; 33:
Perrier S, Houdier S, Domine F, Cabanes A, Legagneux L, Sumner AL, Shepson PB. Formaldehyde in Arctic
snow. Incorporation into ice particles and evolution in the snowpack. Atmospheric Environment 2002;
36: 2695.
Petit JR, Jouzel J, Raynaud D, Barkov NI, Barnola J-M, Basile I, Bender M, Chappellaz J, Davis M, Delaygue
G, Delmotte M, Kotlyakov VM, Legrand M, Lipenkov VY, Lorius C, Pepin L, Ritz C, Saltzman E, Stievenard M. Climate and atmospheric history of the past 420,000 years from the Vostok ice core, Antarctica. Nature 1999; 399: 429.
Ribas-Carbo M, Still C, Berry J. Automated system for simultaneous analysis of 13C, 18O and CO2 concentrations
in small air samples. Rapid Communications In Mass Spectrometry 2002; 16: 339.
Rice AL, Gotoh AA, Ajie HO, Tyler SC. High-precision continuous-flow measurement of δ13C and δD of atmospheric CH4. Analytical Chemistry 2001; 73: 4104.
Riedel K, Allan W, Weller R, Schrems O. Discrepancies between formaldehyde measurements and methane
oxidation model predictions in the Antarctic troposphere: An assessment of other possible formaldehyde sources. Journal of Geophysical Research D 2005; 110: 1.
Röckmann T, Kaiser J, Brenninkmeijer CAM, Crowley JN, Borchers R, Brand WA, Crutzen PJ. Isotopic enrichment of nitrous oxide (15N14NO, 14N15NO, 14N14N18O) in the stratosphere and in the laboratory.
Journal of Geophysical Research D 2001; 106: 10403.
Rommelaere V, Arnaud L, Barnola J-M. Reconstructing recent atmospheric trace gas concentrations from polar
firn and bubbly ice data by inverse methods. Journal of Geophysical Research 1997; 102: 30.
Ruddiman WF. The anthropogenic greenhouse era began thousands of years ago. Climatic Change 2003; 61:
261.
Ruddiman WF. The early anthropogenic hypothesis a year later: An editorial reply. Climatic Change 2005; 69:
427.
Salamatin AN, Lipenkov VY, Ikeda-Fukazawa T, Hondoh T. Kinetics of air-hydrate nucleation in polar ice
sheets. Journal of Crystal Growth 2001; 223: 285.
Santrock J, Studley SA, Hayes JM. Isotopic analyses based on the mass spectrum of carbon dioxide. Analytical
Chemistry 1985; 57: 1444.
Schaefer H. Stable Carbon Isotopic Composition of Methane from Ancient Ice Samples. Dissertation, University
of Victoria, 2005;
126
References
Schlitzer R. Carbon export fluxes in the Southern Ocean: Results from inverse modeling and comparison with
satellite-based estimates. Deep-Sea Research Part II: Topical Studies in Oceanography 2002; 49: 1623.
Schmitt J, Glaser B, Zech W. Amount-dependent isotopic fractionation during compound-specific isotope analysis. Rapid Communications In Mass Spectrometry 2003; 17: 970.
Schwander J, Barnola J-M, Andrié C, Leuenberger M, Ludin A, Raynaud D, Stauffer B. The age of the air in the
firn and the ice at Summit, Greenland. Journal of Geophysical Research 1993; 98: 2831.
Schwander J, Stauffer B, Sigg A. Air mixing in firn and the age of the air at pore close-off. Annals of Glaciology
1988; 10: 141.
Severinghaus JP, Battle MO. Fractionation of gases in polar ice during bubble close-off: New constraints from
firn air Ne, Kr and Xe observations. Earth and Planetary Science Letters 2006; 244: 474.
Severinghaus JP, Bender ML, Keeling RF, Broecker WS. Fractionation of soil gases by diffusion of water vapor,
gravitational settling, and thermal diffusion. Geochimica et Cosmochimica Acta 1996; 60: 1005.
Severinghaus JP, Grachev A, Battle M. Thermal fractionation of air in polar firn by seasonal temperature gradients. Geochemistry, Geophysics, Geosystems 2001; 2: 146.
Severinghaus JP, Grachev A, Luz B, Caillon N. A method for precise measurement of argon 40/36 and krypton/argon ratios in trapped air in polar ice with applications to past firn thickness and abrupt climate
change in Greenland and at Siple Dome, Antarctica. Geochimica et Cosmochimica Acta 2003; 67: 325.
Severinghaus JP, Sowers T, Brook EJ, Alley RB, Bender ML. Timing of abrupt climate change at the end of the
Younger Dryas interval from thermally fractionated gases in polar ice. Nature 1998; 391: 141.
Siegenthaler U. CO2-Konzentrationsmessungen an polaren Eisbohrkernen und Tests mit einer neuen Extraktionsmethode, Diplomarbeit, Bern. unpublished 2002;
Siegenthaler U. Atmosphärische CO2-Messungen der letzten 650.000 Jahre anhand von Messungen an Antarktischen Eisbohrkernen. Dissertation, University of Bern 2006;
Siegenthaler U, Friedli H, Loetscher H, Moor E, Neftel A, Oeschger H, Stauffer B. Stable-isotope ratios and
concentration of CO2 in air from polar ice cores. Annals of Glaciology 1988; 10: 151.
Siegenthaler U, Monnin E, Kawamura K, Spahni R, Schwander J, Stauffer B, Stocker TF, Barnola J-M, Fischer
H. Supporting evidence from the EPICA Dronning Maud Land ice core for atmospheric CO2 changes
during the past millennium. Tellus 2005a; 57: 51.
Siegenthaler U, Stocker TF, Monnin E, Lüthi D, Schwander J, Stauffer B, Raynaud D, Barnola JM, Fischer H,
Masson-Delmotte V, Jouzel J. Atmospheric science: Stable carbon cycle-climate relationship during the
late Pleistocene. Science 2005b; 310: 1313.
Sigman DM, Boyle EA. Glacial/interglacial variations in atmospheric carbon dioxide. Nature 2000; 407: 859.
Smith HJ, Fischer H, Wahlen M, Mastroianni D, Deck B. Dual modes of the carbon cycle since the Last Glacial
Maximum. Nature 1999; 400: 248.
Smith HJ, Wahlen M, Mastroianni D, Taylor K, Mayewski P. The CO2 concentration of air trapped in
Greenland Ice Sheet Project 2 ice formed during periods of rapid climate change. Journal of Geophysical Research 1997a; 102: 26577.
Smith HJ, Wahlen M, Mastroianni D, Taylor KC. The CO2 concentration of air trapped in GISP2 ice from the
Last Glacial Maximum-Holocene transition. Geophysical Research Letters 1997b; 24: 1.
Sowers T, Bender M, Raynaud D, Korotkevich YS. δ15N of N2 in air trapped in polar ice: a tracer of gas transport in the firn and a possible constraint on ice age-gas age differences. Journal of Geophysical Research 1992; 97:
References
127
Sowers T, Bernard S, Aballain O, Chappellaz J, Barnola JM, Marik T. Records of the δ13C of atmospheric CH4
over the last 2 centuries as recorded in Antarctic snow and ice. Global Biogeochemical Cycles 2005; 19:
1.
Spahni R, Chappellaz J, Stocker TF, Loulergue L, Hausammann G, Kawamura K, Flückiger J, Schwander J,
Raynaud D, Masson-Delmotte V, Jouzel J. Atmospheric science: Atmospheric Methane and Nitrous
Oxide of the Late Pleistocene from Antarctic Ice Cores. Science 2005; 310: 1317.
Spahni R, Schwander J, Flückiger J, Stauffer B, Chappellaz J, Raynaud D. The attenuation of last atmospheric
CH4 variations recorded in polar ice cores. Geophysical Research Letters 2003; 30: 25.
Spero HJ, Lea DW. The cause of carbon isotope minimum events on glacial terminations. Science 2002; 296:
522.
Stauffer B, Flückiger J, Monnin E, Nakagawa T, Aoki S. Discussion of the reliability of CO2, CH4 and N2O
records from polar ice cores. Mem. Natl Inst. Polar Res., Spec. Issue 2003; 57: 139.
Stauffer B, Flückiger J, Monnin E, Schwander J, Barnola JM, Chappellaz J. Atmospheric CO2, CH4 and N2O
records over the past 60 000 years based on the comparison of different polar ice cores. Annals of Glaciology 2002; 35: 202.
Stephens BB, Keeling RF. The influence of antarctic sea ice on glacial-interglacial CO2 variations. Nature 2000;
404: 171.
Stevenson BA, Kelly EF, McDonald EV, Busacca AJ. The stable carbon isotope composition of soil organic
carbon and pedogenic carbonates along a bioclimatic gradient in the Palouse region, Washington State,
USA. Geoderma 2005; 124: 37.
Toggweiler JR. Variation of atmospheric CO2 by ventilation of the ocean's deepest water. Paleoceanography
1999; 14: 571.
Traufetter F, Oerter H, Fischer H, Weller R, Miller H. Spatio-temporal variability in volcanic sulphate deposition
over the past 2 kyr in snow pits and firn cores from Amundsenisen, Antarctica. Journal of Glaciology
2004; 50: 137.
Trudinger CM, Enting IG, Etheridge DM, Francey RJ, Levchenko VA, Steele LP, Raynaud D, Arnaud L. Modeling air movement and bubble trapping in firn. Journal of Geophysical Research D: Atmospheres
1997; 102: 6747.
Tschumi J, Stauffer B. Reconstructing past atmospheric CO2 concentration based on ice-core analyses: open
questions due to in situ production of CO2 in the ice. Journal of Glaciology 2000; 46: 45.
Verkouteren RM. Preparation, characterization, and value assignment of carbon dioxide isotopic reference materials: RMs 8562, 8563, and 8564. Analytical Chemistry 1999; 71: 4740.
Wang YQ, Zhang XY, Arimoto R, Cao JJ, Shen ZX. Characteristics of carbonate content and carbon and oxygen
isotopic composition of northern China soil and dust aerosol and its application to tracing dust sources.
Atmospheric Environment 2005; 39: 2631.
Weller R, Traufetter F, Fischer H, Oerter H, Piel C, Miller H. Postdepositional losses of methane sulfonate, nitrate, and chloride at the European Project for Ice Coring in Antarctica deep-drilling site in Dronning
Maud Land, Antarctica. Journal of Geophysical Research D: Atmospheres 2004; 109:
Werner RA, Brand WA. Referencing strategies and techniques in stable isotope ratio analysis. Rapid Communications In Mass Spectrometry 2001; 15: 501.
Werner RA, Rothe M, Brand WA. Extraction of CO2 from air samples for isotopic analysis and limits to ultra
high precision δ18O determination in CO2 gas. Rapid Communications In Mass Spectrometry 2001; 15:
2152.
128
References
Wilson AT, Donahue DJ. AMS carbon-14 dating of ice: progress and future prospects. Nuclear Instruments and
Methods in Physics Research B 1990; 52: 473.
Wilson AT, Donahue DJ. AMS Radiocarbon dating of Ice; Validity of the Technique and the Problem of Cosmogenic in-situ Production in Polar Ice Cores. Radiocarbon 1992; 34: 431.
Wilson AT, Long A. New approaches to CO2 analysis in polar ice cores. Journal of the Geophysical Research
1997; 102: 26601.
Wolff EW, Miners WD, Moore JC, Paren JG. Factors controlling the electrical conductivity of ice from the polar
regions - A summary. Journal of Physical Chemistry B 1997; 101: 6090.
Yeom CK, Lee SH, Lee JM. Study of transport of pure and mixed CO2/N2 gases through polymeric membranes.
Journal of Applied Polymer Science 2000; 78: 179.
Young ED, Galy A, Nagahara H. Kinetic and equilibrium mass-dependent isotope fractionation laws in nature
and their geochemical and cosmochemical significance. Geochimica et Cosmochimica Acta 2002; 66:
1095.
Zhang J, Quay PD, Wilbur DO. Carbon isotope fractionation during gas-water exchange and dissolution of CO2.
Geochimica et Cosmochimica Acta 1995; 59: 107.
Zumbrunn R, Neftel A, Oeschger H. CO2 measurements on 1-cm3 ice samples with IR laserspectrometer (IRLS)
combined with a new dry extraction device. Earth and Planetary Science Letters 1982; 60: 318.
5 Appendix:
Preconditioninung + beginning of tube No. 1
status: molesieve trap +100 °C , V5 closed after 2 min pumping to vacuum via open V6, V7, V8 and V9
0
close V2 and V8, and open V7 and V9
10
immerse CO2 trap in LN
20
open V3, V4, V6
30
remove dewar from sublimation vessel
40
released
air
mount IR lamps
50
passing
switch ventilator on
1
the
set point cold air -110 °C
10
20
increase flow of cold air (~0.2 bar)
CO2 trap
30
IR lamps stage 3
40
is
adjust pH2O to 0.2 -0.3 mbar (PS sensor)
50
pumped
2
IR lamps stage 5
10
increase flow of cold air ( ~0.5 bar)
20
to
open V2 for 10 s
vacuum
30
close V1
40
if sublimation is stable: close V3
50
from now on air released during the sublimation
is finally transferred into tube No 1
close V4 if PH < 0.01 mbar
3
0
[min]
remove LN dewar from CO2 trap and start stopp watch
10
accumulation of
20
30
collected
released
CO2
air during the
pumped
sublimation
40
50
to
1
within the sublimation
vacuum
10
vessel for ~3 min
20
30
40
open V5
turn heater molesieve trap off
close V5
50
immerse molesieve trap in LN
2
10
immerse CO2 trap in LN
20
close V7
30
open V4 and V3
open V8
accumulated air released during the sublimation
is collected
onto the CO2 and molesieve traps
Figure 5-1 Detailed plan of working steps for the sublimation extraction of ice core samples: Part I preconditioning steps and collection of tube No. 1. See general overview of the procedure and used color codes at Figure 3-4.
130
Appendix
Sequence scheme for the succesive collection of 5 sub-samples
during continuous sublimation
close V9
40
immerse glas tube in LN
50
11
close V3
10
close V4 if PH < 0.01 mbar
20
open V7
open V5
close V6 if PM < 0.0020 mbar
30
remove LN dewar from CO2 trap and start stopp watch
40
0
CO2 transfer
turn heater molsieve trap ON
remove LN dewar from molsieve trap
accumulation of
20
ca. 1 min
30
to
40
desorption of
air from
molsieve
air during the
sublimation
tube top
at 100 °C
50
1
released
trap
close V5
within the sublimation
until pressure
equilibrium
vessel for ~3 min
lower LN tube dewar
10
20
read out air pressure at PM sensor
6 min
CO2 transfer
turn heater molsieve trap OFF
close V8
30
open V9
40
open V6 if PM <5 x 10 mbar and
5 repetitions for sub-samples No 1-5
[min]
1 min
10
-4
50
tube top
2
immerse molsieve trap in LN
to
10
immerse CO2 trap in LN
tube tip
close V7
open V4
open V3
20
30
flame seal tube at 2.0 cm
7
accumulated gas released during the
sublimation is collected
onto the CO2 and molesieve traps
open V8 to vacuum
Figure 5-2 Detailed description of the working steps for the sublimation extraction of ice core samples: Part II successive collection of tube No. 1-5. For the general overview of the entire procedure and the used color codes see Figure 3-4.
131
repeat 5 x (EQ-L3, EQ-L2, EQ-L1) , n x (EQ-SA) for n sample tubes, and 4 x (EQ-L3, EQ-L2, EQ-L1)
Appendix
1) LN refill device ON
2) Valco 2 clean mode (direct high He cracker flow to cracker)
3) delay 60 s (flush cracker with high He cracker flow)
4) Valco 2 transfer mode (direct GC carrier to cracker)
5) delay 30 s
6) LN refill device OFF
7) Valco 1 inject mode (direct CO2 reference gas into GC carrier)
8) delay 3 s (empty reference gas loop)
repeat steps
8 to 11
9) Valco 1 fill mode (direct CO2 reference gas to loop)
EQ
3 times
10) delay 4 s (refill reference loop with CO2)
11) delay 36 s
12) actuator cryofocus DOWN
13) delay 90 s (trap CO2 on cryofocus capillary)
14) actuator cryofocus UP (vaporize trapped CO2 and transfer to GC colunm)
15) delay 30 s
16) Valco 2 clean mode (direct high He cracker flow to cracker)
17) delay 80 s (open cracker, remove old tube, insert new tube, close cracker)
18) LN refill device ON
19) delay 60 s (flush atmospheric air out of the cracker)
20) Valco 2 transfer mode (direct GC carrier to cracker)
SA
21) delay 30 s
22) LN refill device OFF
23) delay 50 s (crack tube by bending the cracker)
24) actuator cryofocus DOWN
in case of linearity peaks
L3, L2, L1:
repeat steps 8 to 11 n-times
for n = 3, 2, 1
delay = 36, 43, and 50 s
or
L3, L2, L1
25) delay 90 s (trap CO2 on cryofocus capillary)
26) actuator cryofocus UP (vaporize trapped CO2 and transfer to GC colunm)
27) delay 30 s
Figure 5-3 Sequence of valve operation for the tube-cracker GC IRMS system according to the ISODAT script
written to produce either EQ-SA pairs, in case of the measurement of sample tubes, or EQ-L1 (L2, L3) in case of
linearity pairs to determine the linearity of the tube cracker measurement system (see Fig. 3-2 for the technical
details and Fig. 3-8 for an overview of the entire measurement session).
132
Appendix
d C
Finnigan ISODAT software
13
CO2 concentration
ion current intensity time series:
raw pressure reading PM
ion current time series m44
background correction
temperature correction
find peak maxima of m44
d O correction
using Tex
background correction
m44, m45 and m46
peak detection and integration
17
d C calculation
13
d13C raw data:
pair-std and psample
iair-std and isample
std on/off, EQ, L1, L2, L3, SAair and SAice
trend correction using std on/off and EQ
remove outliers and
calculate CO2 concentration
d C trend corrected data:
13
with Eqn. 3-3
L1, L2, L3, SAair and SAice
MATLAB routine
linearity correction using L1, L2 and L3
d13C linearity corrected data:
SAair and SAice
calculation of whole analysis d C off-set:
13
13
13
assigned d Cair standard - mean d C(SAair)
d13C off-set corrected data:
raw CO2 concentration:
SAice tube No. 1-5
sample tube No. 1-5
outlier removal
outlier removal
calculate mean of tube No. 2-4
calculate mean of tube No. 2-4
gravitation correction
gravitation correction
d13C final value
CO2 concentration final value
Figure 5-4 Calculation chain for δ13C and the CO2 concentration.
133
Appendix
Table 5-1 Measured ice core samples with values for δ13C and CO2 concentration corrected for measurement
biases; values for δ13C-gravity and CO2-gravity were additionally corrected for gravitational settling in the firn
column.
depth
δ13C
δ13C-gravity
CO2
CO2-gravity
[m]
tube No.
[‰ VPDB]
[‰ VPDB]
[ppmv]
[ppmv]
151
150.655
1
-6.74
-7.18
283
281
151
150.655
2
-6.45
-6.89
285
283
151
150.655
3
-6.21
-6.65
281
279
151
150.655
4
-6.17
-6.61
278
276
151
150.655
5
-6.36
-6.80
280
278
151
150.610
1
-6.67
-7.11
288
286
151
150.610
2
-6.28
-6.72
282
281
151
150.610
3
-6.17
-6.61
283
281
151
150.610
4
-6.32
-6.76
285
283
151
150.610
5
-6.28
-6.72
275
273
151
150.565
1
-6.66
-7.10
288
286
151
150.565
2
-6.29
-6.73
276
274
151
150.565
3
-6.03
-6.47
279
277
151
150.565
5
-6.30
-6.74
276
274
253
252.700
1
-6.42
-6.86
276
274
253
252.700
2
-6.17
-6.61
275
273
253
252.700
3
-6.15
-6.59
278
276
253
252.700
4
-6.15
-6.59
280
278
253
252.700
5
-6.15
-6.59
277
275
253
252.655
1
-6.62
-7.06
279
277
253
252.655
2
-6.18
-6.62
277
275
253
252.655
3
-6.07
-6.51
277
275
253
252.655
4
-6.18
-6.62
277
275
253
252.655
5
-6.17
-6.61
267
265
253
252.610
1
-6.70
-7.14
283
281
253
252.610
3
-6.19
-6.63
273
271
253
252.610
4
-6.28
-6.72
276
274
EDML bag No.
to be continued on the next page
134
Appendix
continuation from Table 5-1
depth
δ13C
δ13C-gravity
CO2
CO2-gravity
[m]
tube No.
[‰ VPDB]
[‰ VPDB]
[ppmv]
[ppmv]
253
252.610
5
-6.10
-6.54
271
269
253
252.565
1
-6.48
-6.92
276
274
253
252.565
2
-6.28
-6.72
281
280
253
252.565
3
-6.24
-6.68
282
280
253
252.565
4
-6.20
-6.64
280
278
420
419.700
1
-6.72
-7.16
277
275
420
419.700
2
-6.41
-6.85
275
273
420
419.700
3
-6.42
-6.86
276
274
420
419.700
4
-6.39
-6.83
275
273
420
419.700
5
-6.56
-7.00
274
272
420
419.655
1
-6.70
-7.14
270
268
420
419.655
2
-6.35
-6.79
278
276
420
419.655
3
-6.19
-6.63
275
273
420
419.655
4
-6.36
-6.80
276
274
420
419.655
5
-6.62
-7.06
275
273
420
419.610
1
-6.66
-7.10
274
272
420
419.610
2
-6.40
-6.84
276
274
420
419.610
3
-6.31
-6.75
276
274
420
419.610
4
-6.49
-6.93
272
270
420
419.610
5
-6.51
-6.95
274
272
1056
1055.700
4
-6.24
-6.63
198
197
1056
1055.700
5
-6.53
-6.92
187
186
1056
1055.655
1
-6.41
-6.80
192
191
1056
1055.655
2
-6.23
-6.62
198
197
1056
1055.655
3
-6.17
-6.56
205
203
1056
1055.655
4
-6.30
-6.69
196
195
1056
1055.655
5
-6.36
-6.75
190
189
1056
1055.610
1
-6.50
-6.89
193
192
1056
1055.610
2
-6.23
-6.62
195
193
1056
1055.610
3
-6.18
-6.57
206
205
1056
1055.610
5
-6.39
-6.78
200
199
EDML bag No.
135
Paper:
On the application and interpretation of Keeling plots in paleo
climate research
__
Deciphering δ13C of atmospheric CO2 measured in ice cores
Peter Köhler, Jochen Schmitt, Hubertus Fischer
published in Biogeosciences Discussions, 3, 513-573, 2006
136
On the application and interpretation of Keeling plots in paleo
climate research
—
Deciphering δ 13C of atmospheric CO2 measured in ice cores
Peter Köhler, Jochen Schmitt, Hubertus Fischer
Alfred Wegener Institute, Helmholtz Center for Polar and Marine Research
P.O. Box 12 01 61, D-27515 Bremerhaven, Germany
published in Biogeosciences Discussions, 3, 513-573, 2006
Abstract
The Keeling plot analysis is an interpretation method widely used in terrestrial carbon cycle research to
quantify exchange processes of carbon between terrestrial reservoirs and the atmosphere. Here, we analyse
measured data sets and artificial time series of the partial pressure of atmospheric carbon dioxide (pCO2 ) and of
δ 13 C of CO2 over industrial and glacial/interglacial time scales and investigate to what extent the Keeling plot
methodology can be applied to longer time scales. The artificial time series are simulation results of the global
carbon cycle box model B ICYCLE. Our analysis shows that features seen in pCO2 and δ 13 C during the industrial
period can be interpreted with respect to the Keeling plot. However, only a maximum of approximately half of
the signal can be explained by this method. The signals recorded in ice cores caused by abrupt terrestrial carbon
uptake or release loose information due to air mixing in the firn before bubble enclosure and limited sampling
frequency. For less abrupt changes as occurring during glacial cycles carbon uptake by the ocean cannot longer be
neglected. We introduce an equation for the calculation of the effective isotopic signature of long-term changes
in the carbon cycle, in which the ocean is introduced as third reservoir. This is a paleo extention of the two
reservoir mass balance equations of the Keeling plot approach. Steady state analyses of changes in the terrestrial
and marine biosphere lead to similar effective isotopic signatures (−8.6h) of the carbon fluxes perturbing the
atmosphere. These signatures are more positive than the δ 13 C signals of the sources, e.g. the terrestrial carbon
pools themselves (∼ −25h). In all other cases the effective isotopic signatures are larger (−8.2h to −0.7h),
and very often indistinguishable in the light of the uncertainties. Therefore, a back calculation from well distinct
fluctuations in pCO2 and δ 13 C to identify their origin using the Keeling plot approach seems not possible.
1 Introduction
In carbon cycle research information on the origin of fluxes between different reservoirs as contained in the ratio
of the stable carbon isotopes 13 C/12 C has become more and more important in the past decades. These isotopic
signatures store information about exchange processes because differences in physical properties of atoms and
molecules containing different isotopes of an element lead to isotopic fractionation.
Prominent examples in our context of global carbon cycle research are gas exchange between surface ocean and
atmosphere or photosynthetic production in both the marine and the terrestrial biosphere. Here, the end member of
a carbon flux associated with a given process is in general depleted in the heavier isotope. This is expressed with
the fractionation factor ε of the process, which depends on various environmental parameters such as temperature
or the biological species (see Zeebe and Wolf-Gladrow (2001) for more basic information on carbon isotopes in
seawater).
The isotopic composition of a reservoir is usually expressed in per mil (h) in the so-called ”δ-notation” as the
relative deviation from the isotope ratio of a defined standard (VPDP in the case of δ 13 C):
 [13 C]

[12 C] sample
δ 13 C sample =  [13 C]
[12 C]
standard
− 1 × 103 .
(1)
138
P. Köhler, J. Schmitt, H. Fischer
The fractionation factor ε (in h) between carbon in sample A and in sample B (e.g. before and after some
fractionation step) is related to the δ values by
ε(A−B) =
δA − δB
.
1 + δ B /103
(2)
During photosynthesis the carbon taken up by marine primary producers is typically depleted by On land, the
type of metabolism determines the fractionation factor during terrestrial photosynthesis. C3 plants inhibit a higher
discrimination against the heavy isotope (ε = −15 to −23h) than plants with C4 metabolism (ε = −2 to −8h)
(Mook, 1986).
One prominent interpretation technique of carbon exchange between the atmosphere and other reservoirs,
e.g used in carbon flux studies in terrestrial ecosystems, is plotting the δ 13 C signature of CO2 as a function of
the inverse of the atmospheric carbon dioxide mixing ratio (δ 13 C = f (1/CO2 )). In doing so, the intercept of a
linear regression with the y-axis can under certain conditions be understood as the isotopic signature of the flux,
which alters the content of carbon in the atmospheric reservoir. This approach is called “Keeling plot” after the
very first usage by Charles D. Keeling about 50 years ago (Keeling, 1958, 1961). The application and interpretation of Keeling plots is widely used in terrestrial carbon research and based on some fundamental assumptions
(see review Pataki et al., 2003).
Keeling plots have also been used in paleo climate research in the past years (e.g. Smith et al., 1999; Fischer
et al., 2003), but it seems that the limitations of this approach have not been adequately taken into account so far
to allow for a meaningful interpretation. The aim of this paper therefore is to emphasise what can be learnt from
Keeling plots, if applied on slow, but global processes acting on glacial/interglacial time scales, and to discuss their
limitations. We emphasise what kind of information can be gained from deciphering the δ 13 C signal measured in
ice cores. For this purpose we extent the Keeling plot approach to a three reservoir system and analyse data sets
and artificial time series produced by a global carbon cycle box model, from which we know which processes are
operating.
2 The Keeling plot
The principle of the Keeling plot approach is based on the exchange process of carbon between two reservoirs
and the conservation of mass. Let Cnew be the mass of carbon after the addition of carbon with mass Cadd to an
undisturbed reservoir with mass Cold .
Cnew = Cold + Cadd
(3)
With δ 13 Cx being the carbon isotope signature of the C component x the conservation of mass thus gives us:
Cnew · δ 13 Cnew = Cold · δ 13 Cold + Cadd · δ 13 Cadd
(4)
13
In combining equations 3 and 4 we obtain a relationship between δ Cnew and Cnew :
δ 13 Cnew = Cold · (δ 13 Cold − δ 13 Cadd ) ·
1
+ δ 13 Cadd
Cnew
(5)
Thus, the y-intercept y0 of the linear regression function of equation 5, which describes δ 13 Cnew as a function
of the inverse of the carbon content (1/Cnew ), gives us the isotopic ratio δ 13 Cadd of the carbon added to the
reservoir.
There are two basic assumptions underlying the Keeling plot method: (1) The system consists of a two reservoirs only. (2) The isotopic ratio of the added reservoir does not change during the time of observation. Both
assumptions are in a strict sense rarely fulfilled. Furthermore, there are arguments about which linear regression
model should be used if one assumes measurement errors in both variables (for details see Pataki et al., 2003). The
methodological aspects concerning the choice of a regression model are not the subject of our investigations here.
The Keeling plot approach was used in the past to interprete various different sub-systems of the global carbon
cycle. Keeling (1958, 1961) first used it to identify the contribution of terrestrial plants to the background isotopic
ratio of CO2 in a rural area near the Pacific coast of North America. The component of the terrestrial biosphere
in the seasonal cycle of CO2 over Switzerland (Friedli et al., 1986; Sturm et al., 2005) and Eurasia (Levin et al.,
2002) was investigated with the Keeling plot. The approach was widely used in terrestrial ecosystem research to
identify respiration fluxes (e.g. Flanagan and Ehleringer, 1998; Yakir and Sternberg, 2000; Bowling et al., 2001;
Pataki et al., 2003; Hemming et al., 2005). It was used in paleo climate research within the last years to disentangle
On the application and interpretation of Keeling plots in paleo climate research
-6.0
-6.5
-6.5
-7.5
250
CO2
13
C
TD
30
1
Time [kyr BP]
LD
-8.0
PB
-8.5
1980 2000
Time [yr AD]
o
-7.0
300
200
-6.0
C [ /oo]
CO2 [ppmv]
350
A
13
Time [yr AD]
1000
2000
CO2 [ppmv]
300 250
200
B
.
....... .
-7.5 ......... .
..
.............
-8.0 .....
............
.
-8.5
-7.0
139
Taylor Dome HOL
Taylor Dome GIG
Taylor Dome LGM
Law Dome
Point Barrow
Point Barrow, detrended
o
2
TD HOL: y0 = -9.5 /oo, r = 40%
o
2
TD GIG: y0 = -5.8 /oo, r = 34%
o
2
TD LGM: y0 = -9.5 /oo, r = 48%
o
2
LD ANT: y0 = -13.1 /oo, r = 96%
o
2
PB ORG: y0 = -16.7 /oo, r = 68%
o
2
PB DET: y0 = -25.3 /oo, r = 96%
0.003 0.004 0.005
1/CO2 [1/ppmv]
Figure 1: A compilation of data from Point Barrow PB, the Law Dome LD, and the Taylor Dome TD ice cores
(A: CO2 , δ 13 C; B: Keeling plot). Monthly resolved data (1982 − 2002) from Point Barrow (Keeling and Whorf,
2005; Keeling et al., 2005). Only times where data in both CO2 and δ 13 C were available are considered here.
For the Keeling plot approach the original data (PB ORG) and detrended time series (PB DET) are plotted and
analysed. Data from firn and ice at Law Dome cover the last millenium (Francey et al., 1999; Trudinger et al.,
1999) which includes the anthropogenic rise in CO2 (LD ANT). Data from the Taylor Dome ice core of the last 30
kyr include the glacial/interglacial transition during Termination I (Smith et al., 1999) with the age model of Brook
et al. (2000). Taylor Dome data are split into the Holocene (TD HOL), the glacial/interglacial transition (TD GIG),
and the LGM (TD LGM).
the processes explaining the subtle changes in CO2 during the relatively stable LGM and the Holocene as well as
the approximately 80 ppmv increase from the Last Glacial Maximum (LGM) to the Early Holocene (Smith et al.,
1999; Fischer et al., 2003; Eyer, 2004).
3 Global CO2 and δ 13 C times series of different temporal resolution
It has been shown (Fischer et al., 2003), that the seasonal amplitude of CO2 and δ 13 C during the last decades,
the anthropogenic rise in CO2 and the corresponding decrease in its δ 13 C signal since 1750 AD, and the glacial/
interglacial variation in these two records exhibit significant different behaviour if analysed with the Keeling plot
approach. Data sets showing these dynamics on the three different temporal scales are compiled in Fig. 1. The
seasonal signal is measured from 1982 to 2002 AD at Point Barrow, Alaska (Keeling and Whorf, 2005; Keeling
et al., 2005). For the anthropogenic variation during the last millennium we use the data measured in air enclosures
in the Law Dome ice core (Francey et al., 1999; Trudinger et al., 1999). Glacial/interglacial variations (1 − 30
kyr BP) were detected in the Taylor Dome ice core (Smith et al., 1999). For the interpretation of the seasonal
signal measured at Point Barrow both data sets (CO2 , δ 13 C) need to be detrended to separate the two simultaneous
occurring effects of the anthropogenic CO2 rise and the seasonality from each other. Annual variations are then
analysed as perturbations from the mean values during the first year of the measurements. The component Cadd
in Equations 3–5 reflects the exchange of carbon of an external reservoir (winter time carbon release from the
terrestrial biosphere in the seasonal signal of Point Barrow and anthropogenic emissions in the case of Law Dome)
with the atmospheric reservoir. The Point Barrow and Law Dome data can be approximated consistently with
the typical Keeling plot linear regression function (r2 = 96% in both). The y-axis intercept y0 declines from the
seasonal effects (–25h) to the anthropogenic impact (–13h) with the mixed signal of the untreated data at Point
Barrow in-between (–17h) (Fig. 1B). This decline is explained by a larger oceanic carbon uptake and a smaller
airborne fraction of any atmospheric disturbance in CO2 in longer time scales (Fischer et al., 2003).
These two examples based on accurate data sets are already beyond Keelings original idea as they are no
longer based on a two reservoir system and highlight the limitations of this approach. While the y-intercept of
the detrended data at Point Barrow match the expectations well, the intercept found in the anthropogenic rise in
Law Dome does not record the δ 13 C signal of the carbon released by anthropogenic activity (with δ 13 C of about
−25 to −30h) to the atmosphere anymore. The seasonal amplitude at Point Barrow can be explained with the
-1
9
8
7
6
5
4
3
2
1
0
500
fossil fuel
land use change
sum
fossil fuel
land use change
sum
400
300
200
100
A
1800
1900
Time [yr AD]
2000 1800
1900
Time [yr AD]
B
0
2000
Cumulative carbon flux [PgC]
P. Köhler, J. Schmitt, H. Fischer
Carbon flux [PgC yr ]
140
Figure 2: Fossil fuel emissions since 1750 AD (Marland et al., 2005), land use change since 1850 AD (Houghton,
2003), linearly extrapolated to zero in year 1750 AD. A: Annual fluxes; B: Cumulative fluxes.
seasonality of the terrestrial biosphere. Vegetation grows mainly in the northern hemisphere, and thus CO2 minima
occur during maximum photosynthetic carbon uptake by plants during northern summer. The seasonal fluctuation
in CO2 should therefore bear a δ 13 C signal of the order of −25h which would account for fractionation during
terrestrial photosynthesis (Scholze et al., 2003). This δ 13 C signal of the seasonal cycle is seen in the detrended
Point Barrow data and corresponds well with other studies (e.g. Levin et al., 2002). The anthropogenic rise seen in
the Law Dome data set is the residual of the combination of fossil fuel emissions (Marland et al., 2005), land use
changes (Houghton, 2003), terrestrial carbon sinks due to CO2 fertilisation (Plattner et al., 2002), all from which
the ocean carbon uptake during that time has to be subtracted. Until about 1910 AD the fossil fuel emissions were
smaller than the carbon release caused by land use change (both around 0.8 PgC yr−1 ). The relation changed
thereafter and in the year 2000 fossil fuel emissions were already more than three times larger than carbon fluxes
based on land use change (6.7 versus 2.1 PgC yr−1 ; Fig. 2A). The cumulative release of fossil fuel carbon outcompeted that from land use change only in year 1973 AD. Altogether about 465 Pg of carbon were released by
anthropogenic activities during this 250 year period to the atmosphere (Fig. 2B). Without carbon uptake by the
ocean and the terrestrial reservoirs this would have led to a rise in atmospheric pCO2 by more than 200 ppmv. The
δ 13 C signal of recent fossil fuel emissions in the USA is around −29 to −30h (Blasing et al., 2004), while carbon
fluxes from land use changes bear the typical δ 13 C signal of the terrestrial biosphere (−25h). These anthropogenic
processes can by no means be inferred from the Keeling plot analysis. The reason for this is that due to the gas
exchange between ocean and atmosphere the basic assumption of a two reservoir system is intrinsically violated.
This questions the applicability of the Keeling plot approach to carbon change studies on long time scales, where
this ocean/atmosphere gas exchange becomes even more significant.
Going further back in time, the glacial/interglacial rise in CO2 and its accompanied δ 13 C variations as measured
in the Taylor Dome ice core led to a sub-grouping of the δ 13 C – 1/CO2 data pairs (Smith et al., 1999) with different
linear regression functions for the Last Glacial Maximum (LGM), the glacial-interglacial transition (GIG), and the
Holocene (HOL) (Fig. 1B). With on average one data point every thousand years the data set is sparse. However,
the y-intercepts during LGM and Holocene are similar (−9.5h) and significant different from that during the
transition (−5.8h). Thus, it was hypothesised that underlying processes for variations in CO2 during the relatively
stable climates of the LGM and the Holocene might have been the same and might have been mainly based on
processes concerning the terrestrial biosphere (Smith et al., 1999; Fischer et al., 2003).
4 Extending the Keeling plot approach to a three reservoir system
A first estimate for the effective carbon isotopic signature in the atmosphere due to an injection of terrestrial carbon
into the ocean/atmosphere system can be derived when extending the two reservoirs to a three reservoir system.
Here we assume that ocean circulation remained the same and that an equilibrium between ocean and atmosphere
is achieved.
On the application and interpretation of Keeling plots in paleo climate research
141
We have to extend the carbon and isotopic balance according to
A + O = A0 + O0 + B
(6)
Aδ A + Oδ O = A0 δ0A + O0 δ0O + Bδ B .
(7)
and
where A0 = 600 PgC and O0 = 38, 000 PgC are the reservoir sizes of the atmosphere and ocean, respectively,
before an injection of terrestrial carbon of the size B. A and O are the sizes of the atmospheric and ocean reservoirs
after the injection. The δ0 ’s represent the carbon isotopic signatures of the reservoirs before the injection with
δ0A = −6.5h and δ0O = +1.5h and the δ’s after the injection. The isotopic signature of the terrestrial biosphere
δ B = −25h is assumed to be constant. Note, that during the gas exchange and the dissociation of carbonic acid
in the seawater fractionation occurs (according to Eq. 2) εAO ≈ δ0A − δ0O ≈ δ A − δ O ≈ −8h, which is assumed
to remain constant in time.
For the oceanic uptake of carbon we have to take the buffering effect of the carbonate system in seawater into
account (Zeebe and Wolf-Gladrow, 2001). The ratio between the change in CO2 concentration and the change in
dissolved inorganic carbon (DIC) is described by the Revelle or buffer factor β, which is temperature, alkalinity
and DIC dependent:
d[CO2 ]/[CO2 ]
β :=
.
(8)
dDIC/DIC
Any additional carbon injected into the ocean/atmosphere system will be distributed in the two reservoirs
according to the ratio of the sizes before the injection, i.e.
A − A0
A0
=β
.
O − O0
O0
(9)
Using the carbon balance in equation 6 we can calculate the size of the ocean and atmosphere reservoir after the
injection
βA0 + O0 + B
.
(10)
O=
βA0 /O0 + 1
The Revelle factor β in recent surface waters varies between 8 and 16 (Sabine et al., 2004). For the preindustrial
setting β in the surface ocean boxes of our box model B ICYCLE is on average 11.5, with 9 in equatorial waters and
around 12 in the high latitudes. Note that the average Revelle factor of the surface ocean falls to 10 for the climatic
conditions of the LGM.
The effective carbon signature of the isotopic change in the atmosphere δ ∆A can be estimated according to
δ ∆A =
Aδ A − A0 δ0A
.
A − A0
(11)
Using the carbon isotopic balance in equation 7 and replacing δ O = δ A − εAO we obtain
δ ∆A =
A0 +O0 +B−O
A
A0 +O0 +B (A0 δ0
+ O0 δ0O + Bδ B + εAO O) − A0 δ0A
O0 + B − O
.
(12)
When we insert the values for isotopic signatures and reservoir sizes given above and vary the amount of terrestrial carbon added, the isotopic fractionation during gas exchange εAO or the Revelle factor β we obtain varying
effective isotopic signatures δ ∆A of the change in the atmospheric carbon reservoir as shown in Fig. 3. In a setting
for the preindustrial climate conditions, δ ∆A varies nearly linearly with B between −9h and −10h (Fig. 3A),
reflecting the progressive lightening of the overall ocean/atmosphere system the more isotopically depleted terrestrial carbon is added. Note, that these values are similar to the ones derived by Smith et al. (1999) and Fischer et al.
(2003) both for the Holocene and the LGM from Taylor Dome ice core data and which have been interpreted as
indicative of terrestrial carbon reservoir changes during these periods. However, from the interpretation of other
processes changing the global carbon cycle following in section 5.3 it will become apparent that not only terrestrial
carbon release can produce this kind of signal.
In the special case with a Revelle factor β = 1 (no carbonate buffering), and εAO = 0h, (no isotopic fractionation during air/sea transfer), δ ∆A records correctly the isotopic signature δ B = −25h of the terrestrial carbon
release (Fig. 3B). In all other cases, both the buffering of the ocean and the isotopic fractionation during gas exchange have a significant influence on the calculated δ ∆A with the change in the Revelle factor having the strongest
effect for typical ocean surface conditions.
This reveals three major findings:
142
P. Köhler, J. Schmitt, H. Fischer
o
AO
A o
= -8 /oo, = 11.5, A = 600 PgC
[ /oo] with B = 10 PgC, A = 600 PgC
-9.0
A
-9.4
[-]
A o
[ /oo]
-9.2
-9.6
-9.8
-10.0
0
200 400 600 800 1000
B [PgC]
19
-8.5
17
15
-9
13
-9.5
11
9 -10
7
-12
5 -14
3 -20
1 -30
-12 -10 -8
-8
B
-7.75
-8
-8.25
-8.75
-9
-9.25
-9.75
-11
-13
-15
-25
-6
-4
-2
0
o
AO
[ /oo]
Figure 3: Results of the extended Keeling approach with three reservoirs. Effective isotopic signature of the atmosphere δ ∆A as function of (A) the size of the terrestrial release and (B) the Revelle Factor β and the fractionation
during gas exchange εAO . Other variable as given in the figures. The cross in subfigure B marks the preindustrial
state (β = 11.5, εAO = −8.0h).
1. The isotopic signature δ ∆A is dependent on the amount of carbon injected and the setting of the system
described by the three reservoirs, their isotopic signatures, the fractionation factors, and the Revelle factor.
2. For realistic settings δ ∆A stays between of −8.5 and −10h. The signal which can be detected is therefore
much more enriched in 13 C than the carbon released from the terrestrial biosphere with δ B = −25h, which
was the origin of the perturbation.
∆A
3. There exists a boundary δδC→0
in the effective signature of the isotopic change in atmospheric δ 13 C which is
reached if the amount of carbon released to the atmosphere converges to zero. Note, that δ ∆A is not defined
for B = 0 PgC, because the denominator in Eq. 12 becomes zero. Perturbations in the system will only lead
to variations in δ ∆A from this boundary.
Only for processes which are faster than the equilibration time of the deep ocean with the atmosphere, substantial amounts of isotopic depleted carbon stay in the atmosphere allowing for more negative effective δ ∆A values in
the Keeling plot. The latter is seen e.g. for the seasonal variation in CO2 due to the waxing and waning of the biosphere and to a smaller extent also for the input of isotopically depleted anthropogenic carbon into the atmosphere
which has a typical time scale of decades to centuries.
The two reservoir system is a special case of these calculation for a three reservoir system when β = 1 and
εAO = 0h, i.e. when the atmosphere and the ocean can be treated as one homogeneous reservoir. The effective
carbon isotopic signature δ ∆A based on our theoretical consideration as calculated in this section is comparable
with the y0 of a linear regression in a Keeling plot performed on measured or simulated data sets. However, details
in the marine carbon cycle, such as spatial variations in the Revelle factor, ocean circulation schemes and the ocean
carbon pumps which introduce vertical gradients in DIC and 13 C in the ocean prevent us from a direct comparison
of the obtained values. Nevertheless, the theoretical exercise above gives us valuable insights for the interpretation
of the artificial data sets, which will be discussed in the following.
5 Artificial pCO2 and δ 13 C times series
Recently, a time-dependent modelling approach was proposing a mechanistic understanding of the dynamics of the
atmospheric carbon records over Termination I by forcing the global ocean/atmosphere/biosphere carbon cycle box
model B ICYCLE forward in time (Köhler et al., 2005a). They identified the impacts of different processes acting
on the carbon cycle on glacial/interglacial time scales and proposed a scenario, which provides an explanation
the evolution of pCO2 , δ 13 C, and ∆14 C over time. The results are in line with various other paleo climatic
observations.
In the following we will reanalyse the results of Köhler et al. (2005a) by applying the Keeling plot analysis to
study whether this kind of analysis applied on paleo climatic changes in atmospheric CO2 and δ 13 C can lead to
meaningful results. Additionally, further simulations with the B ICYCLE model will be performed. The advantage
On the application and interpretation of Keeling plots in paleo climate research
143
of using model-generated artificial time series is, that we know which processes are operating and which processdependent isotopic fractionations influence the δ 13 C signals of the results. We highlight how the Keeling plot
approach can gain new insights from these data sets and where its limitations in paleo climatic research seem to
be.
Since B ICYCLE does not resolve seasonal phenomena we are unable to interprete or reconstruct the dynamics
of the Point Barrow data set of the last decades. However, we are able to implement the anthropogenic impacts
of the last 250 years as seen in the Law Dome ice core. After a short model description and an interpretation of
this data set as a sort of ground truthing for our analysis, we dig into the glacial/interglacial mystery of the carbon
cycle and re-evaluate the Taylor Dome ice core data set.
Please note, that atmospheric scientists typically measure carbon dioxide as volume mixing ratio in parts per
million and volume (ppmv) in dry air. Marine chemists and the artificial records produced by our model give
carbon dioxide as partial pressure (pCO2 ) given in units of µatm. Only in dry air and at standard pressure, they are
numerically equal (Zeebe and Wolf-Gladrow, 2001).
5.1
The global carbon cycle box model B ICYCLE
The box model of the global carbon cycle B ICYCLE consists of an ocean module with ten homogeneous boxes
in three basins (Atlantic, Southern Ocean, Indo-Pacific) and three different vertical layers (surface, intermediate,
deep), a globally averaged atmospheric box and a terrestrial module with seven globally averaged compartments
representing ground and tree vegetation and soil carbon with different turnover times (Fig. 4). Prognostic variables
in the model are DIC, alkalinity, oxygen, phosphate and the carbon isotopes 13 C and 14 C in the ocean boxes,
and carbon and its carbon isotopes in the atmosphere and terrestrial reservoirs. The net difference between sedimentation and dissolution of CaCO3 is calculated from variations of the lysocline and imposes fluxes of DIC and
alkalinity between deep ocean and sediment. The model is completely described in Köhler and Fischer (2004) and
Köhler et al. (2005a). B ICYCLE is based in its architecture on earlier box models used during the past two decades
(Emanuel et al., 1984; Munhoven, 1997). It was adapted to be able to answer questions of paleo climate research
with its whole parameterisation being updated.
We apply disturbances of the climate system through the use of forcing functions and paleo climate records
(e.g. changes in temperature, sea level, aeolian dust input in the Southern Ocean) and prescribe changes in ocean
circulation over time based on other data- and model-based studies. B ICYCLE is then able to reconstruct the
evolution of atmospheric pCO2 , δ 13 C, and ∆14 C during the last glacial/interglacial transition (Köhler et al., 2005a).
Atmosphere
50°N
SURFACE
1
100 m
INTER−
MEDIATE
40°S
40°S
9
10
16
3
15
1
C4
1
4
9
C3
30
NW
W
2
D
5
16 16
1000 m
40°N
19
6
12
9
16
5
FS
3
SS
20
Biosphere
DEEP
22
18
6
Atlantic
Rock
9
SO
Indo−Pacific
Sediment
Box model of the Isotopic Carbon cYCLE
11
00
00
11
111
00 000
11
000
111
00
11
000
111
000
111
000
111
000
111
000000
111111
00000
11111
000000
111111
00000
11111
000000
111111
00000
00000011111
111111
00000
11111
000000
111111
00000
11111
000000
00000
11111
water111111
000000
111111
00000
00000011111
111111
00000
11111
carbon
BICYCLE
Figure 4: A sketch of the B ICYCLE model including boundary conditions and preindustrial ocean circulation
fluxes (in Sv = 106 m2 s−1 ) in the ocean module. The globally averaged terrestrial biosphere distinguishes ground
vegetation following different photosynthetic pathways (C4, C3), non-woody (NW), and woody (W) parts of trees,
and soil compartments (D, FS, SS) with different turnover times.
144
P. Köhler, J. Schmitt, H. Fischer
400
-6.5
pCO2 [ atm]
400 350
300
-6.5 13
o
380
-7.0
-7.0
-7.5
-7.5
Cant = -20
/oo2
o
y0 = -11.8 /oo; r =99%
300
o
-8.0
-8.5
-9.0
-9.5
-9.5
-10.0
-10.0
400
-6.5
-6.5
380
-7.0
-7.0
-7.5
-7.5
260
360
340
320
pCO2 TB active
13
o
C, -20 /oo
13
o
C, -25 /oo
13
o
C, -30 /oo
-8.0
-8.5
300
280
260
1500
C
1750
Time [yr AD]
o
A
C [ /oo]
-9.0
280
pCO2 [ atm]
-8.5
C [ /oo]
320
-8.0
13
340
pCO2 TB passive
13
o
C, -20 /oo
13
o
C, -25 /oo
13
o
C, -30 /oo
13
pCO2 [ atm]
13
360
13
B
13
o
Cant o= -202/oo
y0=-9.2 /oo; r =94%
13
13
-9.5
-9.5
-10.0
0.0025
o
Cant =o-25 /2oo
y0=-11.3 /oo; r =97%
-8.5
-9.0
o
Cant = -30
/oo
o
y20 = -15.7 /oo
r =99%
-8.0
-9.0
-10.0
2000
o
Cant = -25
/oo
o
y20 = -13.7 /oo
r =99%
o
Cant =o-30 /2oo
y0=-13.5 /oo; r =98%
D
0.003
0.0035
1/pCO2 [1/ atm]
Figure 5: Reconstructions of the rise in pCO2 during the last 500 years with B ICYCLE (left: pCO2 , δ 13 C; right:
Keeling plot). Anthropogenic fluxes were used as plotted in Fig. 2. Two different settings are tested, one with
passive terrestrial biosphere TB (top), meaning that carbon storage in the land reservoirs was kept constant, and
one with active terrestrial biosphere (bottom), in which a rise in the internal calculated pCO2 is enhancing terrestrial carbon uptake via its fertilisation effect. Different simulations with different isotopic signatures of the
anthropogenic emission (−20h, −25h, −30h).
It was further used to propose a mechanistic understanding of variation in atmospheric CO2 during the last eight
glacial cycles (Wolff et al., 2005; Köhler and Fischer, 2006), and to analyse the implication of changes in the
carbon cycle on atmospheric ∆14 C and on the 14 C production rates (Köhler et al., 2006).
5.2
Anthropogenic emissions — ground truth of the paleo Keeling plot approach
We implement a data-based estimate of the anthropogenic emission since 1750 AD in our model as seen in Fig. 2.
B ICYCLE calculates pCO2 depending on the dynamics of the terrestrial biosphere. In the more realistic case of an
active terrestrial biosphere, implying an enhanced photosynthesis and thus carbon uptake through CO2 fertilisation,
pCO2 at year 2000 is calculated to 351 µatm (Fig. 5C). A scenario with passive terrestrial biosphere, meaning a
constant carbon storage over time, leads to 388 µatm in the same year (Fig. 5A). The annual mean in year 2000 in
the atmospheric CO2 data at Point Barrow is 371 ppmv, which is approximately half way between the results of
the two different simulation scenarios. The scenario with passive terrestrial biosphere is easier to interprete, since
we only have to consider the anthropogenic carbon flux to the atmosphere and the effect of the oceanic sink. Both
scenarios will be analysed in the following.
The precise value of the δ 13 C signature of anthropogenic caused carbon release is still uncertain, e.g. land use
change has a δ 13 C of −25h (Scholze et al., 2003), while δ 13 C of fossil fuel emissions is around −30h (Blasing
et al., 2004). We therefore varied the isotopic signature of the anthropogenic carbon fluxes between −20h and
−30h to evaluate the importance of this signature for the simulation results. The simulated atmospheric δ 13 C in
year 2000 AD was −8.4h, −9.1h, −9.7h and −7.4h, −7.9h, −8.4h in the scenario with passive and active
terrestrial biosphere and for different δ 13 C signatures (−20h, −25h, −30h), respectively (Fig. 5A,C), reflecting
a larger terrestrial fixation of anthropogenic carbon in the active scenario. The annual average δ 13 C measured at
On the application and interpretation of Keeling plots in paleo climate research
145
different globally distributed stations varied between −8.0h and −8.2h (Keeling et al., 2005).
The regression functions of the Keeling approach are still a good approximation of the artificial data sets (r2 ≥
94%, Fig. 5B,D). However, the y-axis intercept varies depending on the assumed δ 13 C signal of the anthropogenic
carbon flux and the mode of the terrestrial biosphere (active/passive) between −9.2h and −15.7h, while the Law
Dome data show −13.1h (Table 1). Note, that these numbers are significantly higher than the isotopic signature
of the anthropogenic carbon added to the system. Due to the non-negligible effect of a third reservoir, the ocean,
the Keeling y-axis intercept deviates from the expected flux signature derived in section 4. Normalised to the
δ 13 C signal of the anthropogenic flux the y-axis intercept amounts to 52 − 59% (passive terrestrial biosphere) and
45 − 46% (active terrestrial biosphere) of the isotopic signal of the anthropogenic flux (Table 1). The difference
to an ideal Keeling plot, in which the whole signal would be explained by the y-axis intercept has to be explained
purely by oceanic uptake in the case of a passive terrestrial biosphere, and by a mixture of terrestrial and oceanic
uptake in simulations with active terrestrial biosphere.
Natural changes in atmospheric CO2 over the past 650,000 years as recorded in Antarctic ice core records (Petit
et al., 1999; Siegenthaler et al., 2005) were always slower and smaller in amplitude than the anthropogenic impact
of the last 250 years. Therefore it is conservative to assume that the oceanic uptake of a terrestrial disturbance in
the past will always be greater than during the anthropogenic period. The potential of the Keeling plot approach to
paleo climate research therefore seems to have an upper limit. No more than about 50% of the isotopic signature
of a carbon source to the atmosphere can be explained with it. In fact, due to the longer time scales on which
most processes act during glacial cycles it can be expected that much less than this upper limit can be explained
by the Keeling plot approach. For steady state situations (the atmosphere and the ocean are in equilibrium) the
∆A
perturbation of the carbon cycle through terrestrial carbon release with a signature of −25h leads to a δδC→0
of
about −9h, as shown in section 4. Thus, for this situations only a fraction of 9/25 = 0.36 is explainable with the
Keeling plot approach.
5.3
Glacial/interglacial times
Besides this upper limit of a signal interpretation due to oceanic carbon uptake in long time series two other
factors make a comparison of artificial time series with long ice core data sets difficult: First, the air which is
enclosed in bubbles in the ice can circulate through the firn down to the depth where bubble close off occurs
Table 1: Analysis of ground truth of the Paleo-Keeling approach: Simulating the anthropogenic rise in pCO2 .
Experiments with different δ 13 Cant signatures of the anthropogenic carbon flux and for two different systems
including an active and a passive terrestrial biosphere are analysed. Displayed are the y-axis intercepts y0 of the
Keeling plots and the ratio which is explained by this approach (y0 / δ 13 Cant ).
δ 13 Cant (h)
Mode of the
terrestrial bioshere
−20
−25
−30
y0 (h)
passive
active
−11.8
−9.2
−13.7
−11.3
−15.7
−13.5
y0 / δ 13 Cant (-)
passive
active
0.59
0.46
0.55
0.45
0.52
0.45
146
P. Köhler, J. Schmitt, H. Fischer
(∼ 70 − 100 m) before it is entrapped in the ice. The bubble close off is a slow process with individual bubbles
closing at different times and depth. Accordingly the air enclosed in bubbles and in an ice sample is subject to a
wide age distribution acting as an efficient low-pass filter on the atmospheric record. Therefore, all information
from the gaseous components of the ice cores is averaged over a time interval of the age of the firn / ice transition
zone. This time interval is depending on temperature and accumulation rates, but can roughly be estimated by the
ratio of the depth of the firn / ice transition zone divided by the accumulation rate (Schwander and Stauffer, 1984).
Thus, the time integral in the gas is small (< 20 years) at Law Dome (Etheridge et al., 1996), varies at Taylor
Dome between 150 years in the Holocene and 300 years in the LGM (Steig et al., 1998a,b), and at EPICA Dome
C, at which the most recent CO2 and δ 13 C measurements were performed (Monnin et al., 2001; Eyer, 2004; Eyer
et al., 2004; Siegenthaler et al., 2005), between 300 and 600 years (Schwander et al., 2001). Second, the CO2 and
δ 13 C records retrieved from ice cores are never continuous records, but consist of single measurements with large,
but un-regular data gaps in between. In the Taylor Dome ice core these gaps are on average approximately 1000
years wide. It will be therefore of interest to investigate if the temporal resolution in the data set will be sufficient
enough to resolve information potentially retrievable through the Keeling approach.
Anthropogenic activities add carbon via land use change and fossil fuel emissions to the atmosphere, and only
subsequently absorbed by the ocean. The causes for natural changes in atmospheric CO2 and thus the carbon
cycle during glacial/interglacial times were to a large extent located in the ocean. Thus, the causes and effects
respectively their timing are in principle different. For the natural glacial/interglacial variations the carbon content
of the atmosphere is determined by the surface ocean, the atmosphere is also called slave to the ocean, while for
the anthropogenic impact the opposite is the case: The carbon of the surface ocean is modified by the injection of
the anthropogenic rise in atmospheric CO2 .
This situation has also consequences for the investigation of different processes causing natural changes in
the carbon cycle. We concentrate in the following on the individual impacts of six important processes. We first
investigate the maximum impacts possible from changes in these processes and then analyse variations of realistic
amplitude. These processes are changes in terrestrial carbon storage, export production of the marine biota, ocean
circulation, gas exchange rates and their variation through variable sea ice cover, and physical effects of variable
sea level and ocean temperature. Please note, that in these factorial scenarios all processes can be treated uncoupled in our box model, e.g. changes in the circulation scheme will not lead to temperature variations, which
might be the case in general circulation models. A summary of this single process analysis is compiled in Table 2.
We end with a combined scenario proposed by Köhler et al. (2005a) which is able the reconstruct the atmospheric
carbon records between 20 and 10 kyr BP. Note, that from these different scenarios only the first one (changes in
terrestrial carbon storage) strictly resembles the initial idea of a Keeling plot (addition/subtraction of carbon from
the atmosphere).
∆A
Table 2: Summary of y-axis intercept y0 and its difference from the terrestrial boundary δδC→0
= −8.4h of the
prior/after Keeling plot analysis for processes changing over Termination I.
Process
y0
(h)
∆A
y0 − δδC→0
(h)
Linear rise in terrestrial carbon storage
Decrease in marine export production
Rise in NADW formation
−8.6
−8.6
−7.8
−0.2
−0.2
+0.6
Rise in Southern Ocean vertical mixing
Decline in sea ice cover / rise in gas exchange rates
Rise in sea level
Rise in temperature
Sediment/ocean interaction
−8.2
−0.7
+0.2
+7.7
−6.4
−3.6
−5.8
+2.0
+4.8
+2.6
Comment
increase was probably non-linear, steepest slope −25h
steeper slope with y0 = −9.7h during first 50 years
varies with time; mixture with changes in marine export
production during Heinrich 1 event; during Younger
Dryas and resumption in the Holocene y0 = −7.15 ±
0.05h, steep slope during first 50 y with y0 = −9.5h
steep slope during first 50 years with y0 = −11.0h
regression over whole data set finds −3.8h with differences in the North (−4.8h) and the South (−77.2h)
On the application and interpretation of Keeling plots in paleo climate research
A
-6.6
-6.6
pCO2 [ atm]
270
260
B
272
-6.8
13
pCO2
13
C
270
-6.7
o
-6.9
266
-7.0
-7.0
-6.5
-6.5
-6.6
-6.6
272
pCO2
13
C
-6.8
-6.9
266
-7.0
-7.0
2
4
Time [kyr]
o
year 1
10 PgC, -23 /oo
o
10 PgC, -13 /oo
o
10 PgC, -33 /oo
o
05 PgC, -23 /oo
D
-6.8
-6.9
0
2
-6.7
268
-2
o
y0 = -18.7 /oo; r =68%
o
-6.7
C [ /oo]
-6.9
C
y0 = -8.4 /oo
-6.8
268
270
o
y0 = -23.8 /oo
o
-6.7
C [ /oo]
year 0
274
pCO2 [ atm]
-6.5
13
pCO2 [ atm]
274
-6.5
147
6
8
10 PgC, -23.o/oo
o
10 PgC, -13 /oo
o
10 PgC, -33 /oo
o
05 PgC, -23 /oo
0.0037
0.00375
1/pCO2 [1/ atm]
Figure 6: Effect of a pulse of instantaneous release (within one year) of terrestrial carbon with an isotopic signature
of δ 13 C = −23h (left: pCO2 , δ 13 C; right: Keeling plot). Different carbon release amplitudes (5, 10 PgC) and
different δ 13 C signatures (−13, −23, −33h) are tested. The regression functions seen in B are for the three
different regression models for the 10 PgC/−23h scenario (red circles). Model 1: year 0 and year 1 (short dash);
model 2: year 0 and year 8,000 (solid); model 3: regression through 375 years (long dash). Large black markers
mark the years 0, 1, 10, 100 in each record. Bottom: Same as above but now the data are smoothed with a 300 yr
running mean.
5.3.1
Terrestrial biosphere
There are two opposing changes in terrestrial carbon storage to be investigated: carbon uptake or carbon release.
Both might happen very fast in the course of abrupt climate anomalies, such as so-called Dansgaard/Oeschger
events (Dansgaard et al., 1982; Johnsen et al., 1992), during which Greenland temperatures rose and dropped by
more than 15 K in a few decades during the last glacial cycle (Lang et al., 1999; Landais et al., 2004). Terrestrial
carbon storage anomalies during these events were estimated with a dynamic global vegetation model to be of the
order of 50 − 100 PgC (Köhler et al., 2005b). The time scales of these anomalies are of the order of centuries
to millenia. We first analyse a scenario in which 10 PgC are released or taken up by the terrestrial pools within
one year. This short time frame of one year was chosen to have experiments, in which the whole carbon flux is
first altering the atmospheric reservoir, before oceanic uptake or release will set in after year one. The amplitude
of the perturbations is optimised to 10 PgC to guarantee still negligible numerical uncertainties (< 0.01h) in the
calculation of the δ 13 C fluxes. We follow with experiments of linear carbon release and three scenarios of carbon
uptake during Termination I to investigate the importance of the time scale for the Keeling plot interpretation.
Fast terrestrial carbon release: This would be the scenario closest to the original Keeling plot analysis in
terrestrial ecosystem research. There is a source (terrestrial biosphere) which emits CO2 directly to the atmosphere.
In this experiment, the 10 PgC release first increases pCO2 by more than 4 µatm immediately after the release, and
equilibrates less than 1 µatm higher than initially (Fig. 6A). The δ 13 C signal shows a drop by more than 0.3h in
year one, and a steady state which is nearly similar to the initial situation (Fig. 6A). Near steady state (±0.1%) in
both pCO2 and δ 13 C was reached 376 years after the carbon release.
There are several possibilities to draw a regression function through the Keeling plot (Fig. 6B):
148
P. Köhler, J. Schmitt, H. Fischer
Table 3: Terrestrial carbon release of different amplitude and isotopic signature and calculated y-axis intercept
based on the original model output, after low-pass filtering of the data with a 300 year running mean, and after data
filtering and reducing the data sets to samples every 100 years.
Amplitude
of the release (PgC)
Scenario
y-axis intercept of different regression models (h)
(r 2 in brackets)
isotopic signature δ 13 Crel
of release (h)
model 1
risinga
model 2
prior — afterb
model 3
equilibrationc
equilibration time (yr)
for model 3
−23.4
−23.4
−33.4
−13.5
−23.8 (100)
−23.6 (100)
−33.9 (100)
−13.6 (100)
−8.4 (100)
−8.4 (100)
−9.3 (100)
−7.5 (100)
−18.7 (68)
−22.3 (75)
−26.0 (68)
−11.4 (70)
375
189
375
375
−23.4
−23.4
−33.4
−13.5
−8.9 (93)
−8.9 (93)
−10.2 (92)
−7.7 (95)
−8.4 (100)
−8.4 (100)
−9.3 (100)
−7.5 (100)
−10.4 (85)
−13.8 (89)
−12.6 (84)
−8.4 (86)
241
71
241
241
−8.6 (100)
−8.6 (100)
−9.6 (100)
−7.5 (97)
−8.4 (100)
−8.4 (100)
−9.3 (100)
−7.5 (100)
−9.1 (100)
not detectabled
−10.5 (100)
−7.7 (100)
150
50
150
150
original model output
10
5
10
10
300 yr running mean
10
5
10
10
300 yr running mean + data selection every 100 yr
10
5
10
10
−23.4
−23.4
−33.4
−13.5
a: This covers data during rise of atmospheric pCO2 , which are only two points in the original data set, but longer
series in smoothed records.
b: Comparing steady state before with new steady state after carbon release.
c: Regression during declining pCO2 in equilibration time.
d: There is only one data point in the time window spanned by the equilibration process, from which no regression
analysis can be performed.
1. A line connecting only the data prior to the start of the carbon release experiment (year 0) and one year later
after 10 PgC are released but before any carbon is taken up by the ocean representing the maximum possible
slope. Thus, pCO2 and δ 13 C after the release can also be calculated following the mass balance equations
of a two reservoir system (Eq. 3 and 4).
2. A straight line through two points characterising the states prior to the carbon release and after re-equilibration.
This would contain the minimum information retrievable in case of low sampling frequency and would be
the analog to the theoretical considerations for a three reservoir system.
3. A regression function through the subset of points covering the equilibration process, in which the main
dynamics of the carbon release are represented. We here choose all points after the release (year 1) until
both pCO2 and δ 13 C were within ±0.1% of their final steady state values.
It is also of interest if and how the amplitude of the carbon release and its isotopic signature influence the
Keeling approach. We therefore performed additional simulations (Fig. 6B,D) with smaller amplitude (5 PgC) and
different δ 13 C signature (−13.5, −23.4, −33.4h). These signatures are the result of the variation of the assumed
global terrestrial fractionation factor εTB = −17h by ±10h.
If such an event of terrestrial carbon release is to be detected in ice cores, we have to manipulate our artificial
data set to account for both the temporal integral during gas enclosures and the limited sampling frequency. We
assumed an average mixing time (running average of 300 years), and a regular sampling frequency of 100 years,
typical for Antarctic ice core studies.
On the application and interpretation of Keeling plots in paleo climate research
149
A summary of calculated y-axis intercepts is found in Table 3. In the original model output the regression
model 1 (analysis of the carbon flux in the year of the release) can explain the δ 13 C signature of terrestrial release
very well, independent of amplitude or the δ 13 C signature itself. The slight overestimation of the regression model
of up to 0.5h might be due to numerical limitations. Differences in the y-axis intercept between scenarios with
varying amplitude and δ 13 C signature of the released carbon are still large in regression model 3 (y0 from −18.7h
to −26.0h), but a simple functional relationship between y-axis intercept y0 and the δ 13 Crel signature of the flux
is missing. In model 2 the different δ 13 C signatures of the carbon release flux are still distinguishable by small
differences in the y-axis intercept (y0 = −7.5, −8.4, −9.3h). These y0 ’s gained from model 2 are similar to the
∆A
boundary δδC→0
introduced in section 4, which is an embedded feature of the system configuration. Interestingly,
the y0 values derived from the B ICYCLE simulations are about 1h isotopically heavier than in our equilibrium
model in section 4. The reason for this is the establishing of vertical gradients in DIC and δ 13 C in the ocean due to
the ocean carbon pumps (Volk and Hoffert, 1985) leading to an enrichment of δ 13 C in the surface water by about
1h.
If we take the signal broadening through temporal mixing in the firn into account (Fig. 6 bottom), the perturbations in pCO2 and δ 13 C are largely reduced to 34% and 7% of their original amplitudes, respectively, and the
duration of the atmospheric pCO2 rise of one year in the original data is now spread over the time length of the
smoothing filter (300 yr). Even for conditions similar to those found at Law Dome where the air is mixed only over
a time interval of 20 years, the amplitudes in pCO2 and δ 13 C are reduces to 70% and 47% of their original values,
respectively. Y-axis intercepts for the 300 years smoothing filter are reduced significantly for regression models 1
(30% to 57% of y0 in original data) and 3 (48% to 74% of y0 in original data).
A further increase of uncertainty arises if we reduce the sampling interval. The effect of a 100 year sampling
frequency reduces y-axis intercepts calculated with regression model 1 and 3 further (Table 3). However, the
uncertainty introduced by reduced sampling frequency is much smaller than the one based on firn air mixing.
∆A
Results obtained with regression model 2 (boundary δδC→0
) were not affected by any of the two post simulation
procedures.
From these fast carbon release experiments, several conclusions can be drawn:
1. The results of regression model 1 are in line with the mass balance equations of the two reservoir Keeling
approach.
2. In all multi-annual experiments the oceanic uptake of carbon will play an important role.
3. Fast terrestrial carbon release events are in their full extent not recordable in the ice core records due to the
time integral introduced by the firn enclosure process.
Slow terrestrial carbon exchange: To understand glacial/interglacial dynamics one has to investigate larger
variations in terrestrial carbon storage of several hundreds of PgC, which occurred over longer time intervals.
We have therefore performed additional experiments, one in which 500 PgC is released by the terrestrial pools,
but now with a constant release rate over a period of 6000 years. In a second set of experiments we mimic in a
Terrestrial C [PgC]
2200
2100
TB0
TB1
TB2
2000
1900
1800
1700
1600
20
18
16
14
12
10
Time [kyr BP]
Figure 7: Changes in the terrestrial carbon storage during the last glacial/interglacial transition follows a nullmodel of linear increase (TB0) or two different dependencies dominated by CO2 fertilisation (TB1) or climate
(TB2). Scenarios taken from Köhler et al. (2005a).
P. Köhler, J. Schmitt, H. Fischer
6
8
A
-6.6
320
-6.7
300
-6.8
-6.7
-6.8
pCO2 [ atm]
300
o
2
o
2
y0 = -8.6 /oo, r = 99%
250
B
280
-6.9
-7.0
-7.0
-6.4
-6.4
-6.5
-6.5
320
-6.6
-6.6
300
-6.7
pCO2 TB0
pCO2 TB1
pCO2 TB2
C
y0 = -8.6 /oo, r = 99%
o
2
y0 = -9.1 /oo, r = 95%
o
2
y0 = -9.0 /oo, r = 94%
D
o
340
C [ /oo]
-6.9
pCO2
13
C
260
pCO2 [ atm]
-6.6
-6.8
280
13
13
260
20
13
18
C TB0
C TB1
C TB2
16 14 12
Time [kyr BP]
13
pCO2 [ atm]
340
-6.5
350
-6.5
o
Time [kyr]
2
4
0
C [ /oo]
-2
13
150
-6.7
-6.8
-6.9
-6.9
-7.0
10
-7.0
TB0
TB1
TB2
0.003
0.0035
0.004
1/pCO2 [1/ atm]
Figure 8: Effects of a gradual change in the carbon storage of the terrestrial biosphere (left: pCO2 , δ 13 C; right:
Keeling plot). Top: Linear decrease in terrestrial carbon storage by 500 PgC. Bottom: Scenarios TB0 (biosphere
only, linear increase in terresterial carbon by 400 PgC between 18 and 11.8 kyr BP) TB1 (dominated by CO2
fertilisation) and TB2 (dominated by climate change) as shown in Fig. 7. Regressions in the Keeling plots are
performed over the whole time period.
simplistic way the carbon uptake of approximately 400 PgC over 6000 yr which might have occurred during the
last glacial/interglacial transition between 18 and 12 kyr BP as assumed in three different scenarios (TB0, TB1,
TB2) in Köhler et al. (2005a). These scenarios differ in functional dependencies of the terrestrial carbon storage
on CO2 fertilisation and climate change (TB0: linear rise in terrestrial carbon; TB1: mainly CO2 dependent; TB2:
mainly climate dependent; Fig. 7).
In both linear experiments (carbon release and TB0) the atmospheric δ 13 C record shows an relaxation behaviour in the first several hundred years after the beginning and after the end of the carbon release with a gradual
change in between (Fig. 8). Atmospheric pCO2 is changing rather constantly over time, also with small nonlinear
responses in the first few hundred years at the beginning and at the end of the experiment. These discontinuities
are caused by the time-delayed oceanic carbon uptake. For example, after the end of the experiment (t = 6 kyr,
Fig. 8A) large parts of the released carbon are taken up by the ocean in the following centuries, similar as in the fast
carbon release experiment shown in Fig. 6. In the more complex scenarios TB1 and TB2 the changes in pCO2 and
δ 13 C are largest in the climate dominated scenarios TB2 with changing rates of up to 30 µatm in pCO2 and 0.3h
in δ 13 C in 1000 years. In the Keeling plots the relaxation behaviour at the beginning and the end of the linear
experiments leads to offsets from the well defined linear relationship. A regression over the whole time period
∆A
leads to a y0 = −8.6h, only 0.2h smaller than the δδC→0
boundary for this system. The scatter of the data
points is larger in TB1 and TB2, but the regression model through the data is still very good (r2 ≥ 94%) leading
to y0 = −9.1 and −9.0h, respectively. The slope of the regression is steeper here, because the fractionation
factor of the terrestrial biosphere εTB is changing over time. The fraction of terrestrial carbon produced by C4
photosynthesis is decreasing from ∼ 30% during the LGM to ∼ 20% during preindustrial times in the scenarios
TB1 and TB2 (Köhler and Fischer, 2004). This leads to a terrestrial fractionation which is more than 1h more
negative in the preindustrial times than in the LGM.
The range of the δ 13 C values concluded from our simulation results for a slow terrestrial carbon release agrees
On the application and interpretation of Keeling plots in paleo climate research
500
500
-6.6
pCO2 [ atm]
400
300
o
y0 = -10.2 /oo
o
-7.0
pCO2
13
C
-7.2
-6.8
year 8000
o
y0 = -8.0 /oo
2
r = 93%
year 100
-7.0
-7.2
13
350
o
-6.8
400
C (atm) [ /oo]
y0 = -8.8 /oo
450
pCO2 [ atm]
-6.6
151
A
300
-7.4
-7.4
-7.6
-7.6
0.002
year 0
250
-2
0
2
4
Time [kyr]
6
8
B
0.003
0.004
1/pCO2 [1/ atm]
Figure 9: Switching from an abiotic to a biotic ocean at t = 0 kyr (A: pCO2 , δ 13 C; B: Keeling plot). Different
regression models in B: model 1 (first year only) in green; model 2 (prior/after) in black; model 3 (equilibration
time) in magenta.
well with the ones proposed by our theoretical three reservoir approach in section 4 and is very different from the
∆A
original δ 13 C of the carbon source. It is especially remarkable that the differences of the y0 ’s from δδC→0
are
very small. Furthermore, these slow carbon exchange processes are so slow that the air enclosure procedure with
the assumed smoothing filter of 300 years would only marginally alter the records and would not change the y0
values. Similarly the restricted sampling frequency is of no importance here. These two processes will therefore
not be analysed any further in the following, because it is reasonable to assume that their impact on the observed
processes can be neglected.
5.3.2
Marine biosphere
While the marine biosphere is in principle a reservoir separate from DIC in the ocean and the atmospheric carbon,
it is not independent because the marine export production establishes vertical gradients in DIC and δ 13 C between
the surface and the deep ocean. Accordingly, the following discussion of changes caused by the marine biosphere
and other factors later-on represents already a misuse of the Keeling plot approach. Nevertheless it is instructive to
study whether the end member analysis can lead to meaningful results and is able to distinguish between different
processes.
For the marine biota we again first want to explore the range of possible results before we analyse one scenario
which seems to be realistic for the last glacial/interglacial transition. We therefore concentrate first on a switch
from an abiotic ocean without any marine biological productivity and no export production to a biotic ocean in
year 0 and vice versa. After these biotic/abiotic experiments, the possible effect of an extended glacial marine
productivity due to the iron fertilisation in the Southern Ocean is explored. In the biotic ocean a flux of 10 PgC
yr−1 of organic carbon is exported at 100 m water depth to the deeper ocean. This organic export production is
coupled via the rain ratio to an export of 1 PgC yr−1 of inorganic CaCO3 .
The abiotic/biotic switch leads to a decrease in atmospheric pCO2 of about 220 µatm and a rise in atmospheric
δ 13 C of 1.0h (Fig. 9A) and the opposite signals in the biotic/abiotic experiment (not shown). The iron fertilisation
experiment decreases glacial pCO2 by 20 µatm, in parallel with a 0.15h rise in δ 13 C (Fig. 10A). Here, both
atmospheric records are relaxing to their preindustrial values after the onset of iron limitation around 18 kyr BP. The
Keeling plot analysis leads to y-axis intercepts of −8.0 to −10.2h for the three different regression models in the
abiotic/biotic experiment (Fig. 9B). Comparing only the prior/after model for all three experiments (abiotic/biotic,
biotic/abiotic, iron fertilisation) (Fig. 9B, 10B) gives nearly identical results (y0 = −8.7 ± 0.1h). If the time
window of analysis is reduced to the first 50 years after the beginning of the reduction in export production in the
iron fertilisation experiment a steeper slope in the Keeling plot leads to y0 of −9.7h (Fig. 10B).
The marine export production combines two of the three ocean carbon pumps: the organic or soft-tissue pump
and the carbonate pump (Volk and Hoffert, 1985). The third one, the solubility pump, operates by the increased
solubility of CO2 in downwelling cold water. They all introduce vertical gradients in DIC in the water column,
the biological pumps additionally build up a gradient in δ 13 C. DIC is reduced in the surface layers through marine
152
P. Köhler, J. Schmitt, H. Fischer
270
-6.3
-6.3
270
pCO2 [ atm]
250
o
-6.5
250
A
18
16 14 12
Time [kyr BP]
-6.4
o
2
50 y: y0 = -9.7 /oo, r =99%
20 kyr BP
o
pCO2
13
C
240
20
C [ /oo]
-6.4
260
13
pCO2 [ atm]
y0 = -8.6 /oo
-6.5
10 kyr BP
-6.6
10
-6.6
0.00375
0.004
1/pCO2 [1/ atm]
B
Figure 10: Effects of iron fertilisation in the Southern Ocean on atmospheric carbon records with scenario taken
from Köhler et al. (2005a) (A: pCO2 , δ 13 C; B: Keeling plot). Blue dots and regression in B over the first 50 years
during pCO2 rise only.
production of both organic material (soft-tissues) and CaCO3 and increased in the abyss through carbon released
during remineralisation and dissolution. During photosynthesis δ 13 C is depleted by about −20h, thus leaving
carbon enriched in 13 C at the surface, while the exported organic matter is depleted. The isotopic fractionation
during the production of hard shells slightly enriches 13 C in the carbonate (ε ∈ [0, 3]h). The vertical gradient in
δ 13 C leads to a difference of about 1.0h between surface (1.5h) and abyss (0.5h) in the biotic ocean, while the
δ 13 C signal in the abiotic ocean does not change with depth and is around 0.55h. If we now switch on the marine
production in a formerly abiotic ocean we merely introduce these gradients to the system. Surface δ 13 C is rising
by 1.0h and so is the atmospheric δ 13 C. The signal seen in the atmospheric record is therefore a mixture of the
fractionation during gas exchange and an increased carbon flux from the atmosphere to the ocean. In B ICYCLE
the flux of CO2 from the surface ocean to the atmosphere has a fractionation factor of εO2A ≈ −10.4h and
εA2O ≈ −2.4h in the opposite direction leading to a net fractionation effect of −8.0h (but both depend also on
2−
temperature, and εO2A additionally on DIC, HCO−
3 , and CO3 ). Similar as in the previous case of a terrestrial
carbon release the system contains a boundary in the effective isotopic signature. Each perturbation of the system
leads to a derivation from this boundary. The effects of changes in the marine carbon fluxes on atmospheric pCO2
∆A
and δ 13 C are not necessary the same as for the terrestrial case. Therefore, the boundary is not identical with δδC→0
,
but seems to be very close. From variations of the global export production one can estimate this marine boundary
to be around −8.5h. In year 1, for example, one can understand the signal (y0 = −10.2h) in the following
way: In areas in which marine export production is reducing surface DIC the gross carbon flux from the ocean
to the atmosphere is largely reduced. In the most extreme case we would only find a gross carbon flux from the
atmosphere to the ocean with the corresponding fractionation factor εA2O = −2.4h. This would be the isotopic
signature of the process in action and added to the marine boundary it would lead at maximum to a y0 of −10.9h.
The calculated y0 is smaller because there is still a small but not negligible gross flux of CO2 from the ocean
to the atmosphere. The signal during the first 50 years of the iron fertilisation experiment (y0 = −9.7h) can
be interpreted similarly. During the latter part of the abiotic/biotic switch experiment and over the equilibration
period y0 is ∼ −8.0h and thus more positive than the marine boundary. This might be caused by the increased
δ 13 C of the DIC in the surface waters. After 100 years, pCO2 has already dropped to 346 µatm, thus nearly 2/3 of
the oceanic uptake of carbon happens in this first century. Therefore the carbon fluxes from the atmosphere to the
ocean and vice versa are nearly similar thereafter. That means that now the isotopic enriched DIC of the surface
waters can enter the atmosphere and is then enriching δ 13 C and y0 .
If compared with the terrestrial experiments the results from regression model 3 (prior/after analysis) have y0 ’s
∆A
which are only 0.2 − 0.4h more negative than δδC→0
. This is very similar to the experiments with slow carbon
exchange between the terrestrial biosphere and the atmosphere.
On the application and interpretation of Keeling plots in paleo climate research
-6.3
270
260
2 3 4
-6.4
pCO2 [ atm]
250
250
-6.5
y0 = -7.8 /oo
o
2
1: y0 = -5.3 /oo, r =52%
o
2
2: y0 = -6.1 /oo, r =36%
o
2
3: y0 = -7.1 /oo, r =90%
o
2
4: y0 = -7.2 /oo, r =88%
1
B
15 kyr BP
11.8 / 16.5 kyr BP
-6.5
240
3
2
13 kyr BP
10 kyr BP
pCO2
13
C
A
230
270
-6.6
-6.6
-6.3
-6.3
(50y): y
4 52 (50y):
y
o
0
0
17 kyr BP
-6.5
240
C
18
16 14 12
Time [kyr BP]
o
-6.4
C (atm) [ /oo]
pCO2
13
C
13
pCO2 [ atm]
y0 = -8.2 /oo
250
230
20
2
= -39.2 /oo, r =85%
o
2
= -9.5 /oo, r =98%
o
260
230
o
o
-6.4
C [ /oo]
1
13
pCO2 [ atm]
270
-6.3
153
-6.4
12 kyr BP
o
50y: y0 = -11.0 /oo
-6.5
2
r =98%
10 kyr BP
-6.6
10
-6.6
D
0.0038 0.004 0.0042
1/pCO2 [1/ atm]
Figure 11: Simulated effects of changes in the strength of the NADW formation (top) and Southern Ocean vertical
mixing (bottom) as used in Köhler et al. (2005a) (left: pCO2 , δ 13 C; right: Keeling plot). Four different time intervals with different strength in NADW formation are identified and labeled in A and B. Regression are performed
for the prior/after situation (black, B, D), for these four time intervals (B) and for selected 50 year time windows
(B, D).
5.3.3
Ocean circulation
Previously Köhler et al. (2005a) assumed a rise in the strength of the North Atlantic Deep Water (NADW) formation from 10 Sv (106 m3 s−1 ) during the Last Glacial Maximum (LGM) to intermediate levels of 13 Sv in the
Bølling/Allerød warm interval and 16 Sv at the beginning of the Holocene. This rise was punctuated by sharp drops
in the NADW formation strength and the subsequent ocean circulation fluxes to 0 Sv and 11 Sv during the Heinrich
1 event and the Younger Dryas, respectively. Repeating this temporal sequence of events over a period of approximately 6000 years leads to drops in pCO2 by 10 µatm during times of reduced ocean circulation accompanied by
rises in δ 13 C of about 0.05h. Initial and final values differ by about 15 µatm and −0.05h (Fig. 11A).
The prior/after analysis in the Keeling plot leads to a y-axis intercept of −7.8h, however, the pattern is highly
time-dependent and allows a breakdown of the time series into the individual events, which show distinctively
different behaviour. These events are marked in Fig. 11A,B (1: decrease in NADW formation during Heinrich 1
event; 2: increase in NADW formation during Bølling/Allerød warm interval; 3: decrease in NADW formation
during Younger Dryas; 4: increase in NADW formation towards interglacial levels). The regression analysis over
the whole of these four periods, which last between 1200 and 2000 years each, finds y-axis intercepts between
−5.3 and −7.2h. If shorter time windows after the beginning of these changes in ocean circulation are analysed,
much steeper regression functions can be found. For example, the regressions through 50 year time windows at
the beginning of interval 2 and 4, which show the steepest slopes in the Keeling plot, lead to y-axis intercepts of
−39.1 and −9.5h, respectively (Fig. 11B).
The complete shut-down of the NADW formation during Heinrich 1 event alters also the nutrient availability
for the marine biota. The export of organic matter depends on the availability of macro-nutrients in the surface
waters and is prescribed to an upper limit of 10 PgC yr−1 . In the time interval between 16.5 and 15 kyr BP marine
export falls from 10 PgC yr−1 to 8.9 − 9.3 PgC yr−1 . Less export production increases atmospheric pCO2 and
154
P. Köhler, J. Schmitt, H. Fischer
decreases atmospheric δ 13 C. This implies that the peaks in the atmospheric carbon records were both dampened
during interval 1. Accordingly, the y0 derived from the Keeling plot during this time interval is a mixture of ocean
circulation and marine biota.
A second ocean circulation process which changed according to Köhler et al. (2005a) over the time of the
last transition was the Southern Ocean vertical mixing rate (Fig. 11C,D). It rose from 9 Sv (glacial) to 29 Sv
(preindustrial) and led to a rise in pCO2 by about 30 µatm and a drop in δ 13 C by more than 0.2h. Here, the
Keeling plot interpretation gives us a y-axis intercept of −8.2h for the prior/after analysis. The steepest slope
during the first 50 years after the start of the change would yield to a y-axis intercept of −11.0h (Fig. 11D).
The overturning circulation distributes carbon in the ocean. Its effect on the atmospheric carbon reservoirs,
however, is opposite to that of the three ocean carbon pumps. While the pumps introduce vertical gradients in
DIC and δ 13 C as described in the subsection about the marine biota, the overturning circulation is reducing these
vertical gradients again through the ventilation of the deep ocean which brings water rich in DIC and depleted
in 13 C back to the surface. A weakening of the ventilation reduces these upwelling processes and leads to lower
pCO2 and higher δ 13 C values as seen in the experiments.
The y0 -values derived from the four intervals for changing NADW formation differ. Between interval 3 and 4
in which opposing changes in ocean circulation occur, the differences in the y0 -values are small. These different
circulation patterns can be seen in the Keeling plot in the dynamics of the first years of the intervals: In intervals
1 and 3, in which the strength of the NADW formation is reduced, the analysis of a time window at the beginning
of the interval leads to less negative y0 than over the whole period, while the opposite is the case in the intervals
2 and 4 with a resumption of the NADW formation. These temporal changes over the course of each interval are
caused by the equilibration of the model to a new steady state. In interval 1, the shutdown of the NADW and the
subsequent fluxes lead first to an enrichment of DIC and of 13 C in the North Atlantic surface waters and thus to
higher atmospheric δ 13 C. Later-on, this is over-compensated by decreasing DIC and δ 13 C in the equatorial surface
of the Atlantic Ocean. However, the dynamics during a complete shutdown of the NADW formation might be
unrealistic because in the current model configuration the tropical Atlantic surface and intermediate ocean boxes
would exchange water with each other but not with any other water masses (Fig. 4). This artefact is also responsible
for the dynamics during the first years of interval 2, but is not affecting the latter intervals.
Compared to the terrestrial carbon release we can say that a change in ocean circulation is assigned to a y0
∆A
slightly more positive than the terrestrial boundary δδC→0
(+0.2 to 1.2h). The dynamics during the first years
of a abrupt rise/decrease in the strength of the overturning circulation lead to larger offsets (fall/rise) from the
terrestrial boundary.
5.3.4
Gas exchange / sea ice
A change in sea ice cover leads to changes in the gas exchange rates with opposing effects for the northern and the
southern high latitudes. As the preindustrial North Atlantic Ocean is a sink for CO2 a reduced gas-exchange due
to higher sea ice cover leads to rising atmospheric pCO2 , while the same happening in the Southern Ocean being
a source for CO2 leads to a drop in pCO2 . Again, we first explore the maximum amplitudes possible by covering
either the whole North Atlantic Ocean or Southern Ocean surface boxes with sea ice, reducing the gas exchange
rates in these areas to zero (Fig. 12 top) before we investigate a data-based scenario for Termination I in which sea
ice was approximately doubled during glacial times (Fig. 12 bottom).
In the extreme scenarios sea ice cover is relaxed instantaneously from a full coverage of the surface ocean
boxes to preindustrial areal distribution in year 0. The experiment in the North Atlantic yields a drop in pCO2
by 35 µatm, a drop in δ 13 C by 0.04h, and a y-axis intercept (prior/after analysis) in the Keeling plot analysis
of −6.1h. A similar experiment in the Southern Ocean increases pCO2 by about 15 µatm, δ 13 C drops by 1.0h
leading to a prior/after y-axis intercept of −23.8h.
In the data based scenario across Termination I the annual averaged sea ice area during LGM is approximately
doubled (Crosta et al., 1998a,b; Sarnthein et al., 2003; Gersonde et al., 2005) and its evolution is coupled to sea
surface temperature changes. pCO2 fluctuates by about 10 µatm and δ 13 C by 0.15h (Fig. 12C). Due to the
different responses in the North and in the South the data plotted as Keeling plot zig-zag quite a lot. Prior/after
analysis finds an y-axis intercept of −0.7h. The regression over the whole period finds y0 = −3.8h (r2 = 38%).
The data based contribution from the North (y0 = −4.8h) and the South (y0 = −77.2h) have very different
y-axis intercepts (Fig. 12D).
The very negative y0 caused by changes in sea ice coverage in the Southern Ocean needs further clarification.
The changes in sea ice coverage cause variations in the gas exchange rates. In the scenario across Termination I,
the global exchange flux rises from 52 PgC yr−1 during the LGM to 58 PgC yr−1 at t = 10 kyr BP in the scenario,
which considers only changes in sea ice in the Southern Ocean. The pCO2 of the atmosphere and the Southern
On the application and interpretation of Keeling plots in paleo climate research
300
-5.6
-5.8
-5.8
-6.0
280
270
-6.2
260
-6.4
250
-2
0
2
4
Time [kyr]
6
8
o
290
-5.6
C [ /oo]
300
pCO2 [ atm]
A
pCO2 N only
pCO2 S only
13
C N only
13
C S only
13
310
pCO2 [ atm]
280
260
year 0
N only
S only
B
o
y0 = -23.8 /oo
-6.0
-6.2
-6.4
-6.6
155
-6.6
o
y0 = -6.1 /oo
year 8000
year 0
0.0033 0.0036 0.0039
1/pCO2 [1/ atm]
pCO2 [ atm]
-6.3
C
-6.4
-6.4
270
-6.5
-6.5
265
260
20
280
260
N&S
N only
S only
D
20 kyr BP
o
275
-6.3
C [ /oo]
pCO2 [ atm]
280
pCO2 N & S
pCO2 N only
pCO2 S only
13
CN&S
13
C N only
13
C S only
13
285
18
16 14 12
Time [kyr BP]
-6.6
10
-6.6
o
y0 = -00.7 /oo
o
2
y0 = -03.8 /oo, r = 38%
o
2
y0 = -04.8 /oo, r = 97%
o
2
y0 = -77.2 /oo, r =100%
10 kyr BP
0.0036
0.0038
1/pCO2 [1/ atm]
Figure 12: Simulated effects of an instantaneous change in sea ice cover from total coverage to preindustrial values
in the North Atlantic (top: N only), or the Southern Ocean (top: S only) and as in the scenario used in Köhler et al.
(2005a) (left: pCO2 , δ 13 C; right: Keeling plot). In the Termination I scenario (bottom) the contribution of northern
(N only) and southern (S only) sea ice is shown individually and in a combined scenario (N & S). Regressions for
the prior/after situation are shown everywhere, in D regressions over the whole period are shown additionally.
Ocean surface box are very close to each other at the beginning of the experiment (270 and 266 µatm, respectively).
Therefore, an increase in gas exchange rate in the Southern Ocean is only marginally increasing atmospheric pCO2
(< 1 µatm). However, the stronger gas exchange leads to a significant decrease in atmospheric δ 13 C by 0.08h. A
y0 of −77.2h is consistent in our modelling environment, but dependent strongly on the model architecture. In
general, the effects of sea ice coverage on carbon cycle dynamics have been found to vary between different models
(Archer et al., 2003). This example shows, that the Keeling plot is difficult to interpret for experiments without
significant net carbon uptake or release of the atmosphere, which lead nevertheless to changes in atmospheric δ 13 C.
The slope of the regression function might in the most extreme case of a constant pCO2 rise to infinity. Besides gas
exchange a second example for this situation is a change from C3 grasses to C4 grasses in the terrestrial biosphere.
∆A
The difference of these results from δδC→0
gives us a comparison with the terrestrial release scenario. The
∆A
contribution from changing gas exchange in the Arctic carries a y0 several h heavier than δδC→0
, while fast
changes in the gas exchange in the South lead to lighter y0 .
5.3.5
Sea level and ocean temperature
Changes in sea level and ocean temperature over the last glacial/interglacial transition are documented rather well
in the paleo climate archives. We therefore describe in the following only those changes which we propose over
Termination I, but do not analyse the maximum range possible by these two processes.
156
P. Köhler, J. Schmitt, H. Fischer
285
-6.5
-6.5
-6.52
-6.52
pCO2 [ atm]
280
-6.56
-6.58
10 kyr BP
o
16 14 12
Time [kyr BP]
pCO2
13
C
C
-6.6
10
-6.6
2
0.0036
0.0037
1/pCO2 [1/ atm]
-6.5
270
-6.5
-6.6
-6.6
pCO2 [ atm]
250
-6.7
-6.8
-6.7
-6.8
230
D
10 kyr BP
o
18
o
y0 = -6.4 /oo, r = 96%
-6.58
C [ /oo]
pCO2 [ atm]
270
265
260
255
250
245
240
235
230
20
-6.54
B
20 kyr BP
y0 = -6.4 /oo
pCO2
13
C
270
20
o
-6.56
275
C [ /oo]
-6.54
13
280
13
pCO2 [ atm]
A
270
o
y0 = -3.6 /oo
o
2
y0 = -3.5 /oo, r =98%
-6.9
-6.9
-7.0
10
-7.0
20 kyr BP
18
16 14 12
Time [kyr BP]
0.0038 0.004 0.0042
1/pCO2 [1/ atm]
Figure 13: Simulated effects of sea level rise (top) and ocean temperature change (bottom) as used in Köhler et al.
(2005a) (left: pCO2 , δ 13 C; right: Keeling plot). Regression of the prior/after analysis (black) and over the whole
period (red) are shown.
Sea level: Sea level rose from 20 kyr BP to 10 kyr BP by about 85 m (Fairbanks, 1990). This leads in our
model to a drop in pCO2 by about 13 µatm and nearly no change in the δ 13 C. The y-axis intercept in a Keeling
plot with prior/after analysis or regression over whole time period is −6.4h (Fig. 13 top).
The rising sea level leads to a dilution of the concentration of all species in the ocean and a decrease in salinity
by about 2.3%. The ocean can then store more carbon. Again, in the comparison with the terrestrial case this sea
∆A
level rise is a process with a typical δ 13 C signature that is about +2h more positive than δδC→0
.
Ocean temperature: The rise of the ocean temperature by 3 to 5 K during the simulated 10 kyr (20 to 10 kyr
BP) leads to a rise in pCO2 of 32 µatm, and a rise in δ 13 C of 0.4h, the latter leading the first by about 1000 years
(Fig. 13 bottom). The Keeling plot analysis gives us y0 ’s of −3.5 and −3.6h for the regression over the whole
period and the prior/after analysis, respectively.
Due to the temperature dependent solubility of CO2 warm water stores less carbon than cold water. A rise in
ocean temperature therefore weakens the solubility pump and leads to an out-gassing of CO2 . The rise in δ 13 C
is mainly caused by the temperature-dependent isotopic fractionation during gas exchange between the surface
ocean and the atmosphere. The temperature effect has a δ 13 C signature which is about +5h heavier than the
∆A
δδC→0
obtained in the terrestrial release case.
5.3.6
Combined scenarios and CaCO3 chemistry
The previous study (Köhler et al., 2005a) finds that combined scenarios are able to reconstruct the observed dynamics in atmospheric pCO2 and δ 13 C during Termination I (Fig. 15). They combine the processes investigated
above (terrestrial carbon storage, marine biology, ocean circulation, temperature, sea level, sea ice) together with
On the application and interpretation of Keeling plots in paleo climate research
C corr:
o
y0 = -5.2 /oo
o
2
y0 = -5.2 /oo, r =99%
-6.6
o
C [ /oo]
-6.6
260
250
-6.7
13
C orig:
o
-6.7
y0 = -5.8 /oo
o
2
y0 = -5.8 /oo, r =99%
20 kyr BP
240
230
20
13
10 kyr BP
13
pCO2 [ atm]
270
pCO2
13
C orig
13
C corr
pCO2 [ atm]
260
240
280
-6.5
-6.5
280
157
A
18
16 14 12
Time [kyr BP]
-6.8
10
-6.8
B
0.00375 0.004 0.00425
1/pCO2 [1/ atm]
Figure 14: Estimation of the effect of sediment/ocean fluxes of CaCO3 on the atmospheric carbon records by
subtracting the results of a scenario with all processes but the CaCO3 effect in operation from the combined
scenario. Terrestrial biosphere is following a linear glacial/interglacial rise (TB0 in Fig. 7). A: pCO2 and δ 13 C. B:
Keeling plot. Regression of the prior/after analysis (black) and over the whole period (blue, green) are shown. The
original δ 13 C (orig) and δ 13 C corrected (corr) for the trend in the mean δ 13 C caused by the sedimentation losses
are shown.
the consideration of CaCO3 fluxes between sediment and ocean.
The influence of the fluxes of CaCO3 cannot be analysed separately as CaCO3 fluxes between deep ocean
and sediment are generated by the model as response to changes in the deep ocean CO2−
3 concentration. CaCO3
chemistry is acting as an amplifier of the results of all other processes with respect to the changes in pCO2 .
However, we can estimate their impact by subtracting the results of a simulation which includes all processes apart
from sediment/ocean fluxes from the results of a scenario including all processes. Over the course of Termination I
the sedimentation of CaCO3 is higher than its dissolution, leading to a loss of DIC (∼1000 PgC) and twice as much
alkalinity. These changes lower the pH of the ocean, which shifts the distribution of the three different species
2−
of DIC (CO2 , HCO−
3 , CO3 ) in the carbonate system towards CO2 , and leads to its out-gassing and a rise in
atmospheric pCO2 of approximately 35 µatm (Fig. 14A). This loss of carbon is accompanied by a depletion of the
mean oceanic δ 13 C by about 0.1h, because the CaCO3 , which is lost to the sediments bears a δ 13 C signal of about
3h. Atmospheric δ 13 C is therefore falling by approximately the same amount. If the δ 13 C record is corrected for
this effect of sedimentation, it rises by about 0.2h (Fig. 14). This residual rise is caused by the fractionation factor
of the CO2 flux from the ocean to atmosphere, which depends on DIC itself and the pH dependent dissociation of
DIC into the different species (Ridgwell, 2001). The Keeling plot analysis of these changes (Fig. 14B) leads to
y0 of −5.8 and −5.2h for both the prior/after analysis and the regression model of the original and the corrected
data, respectively.
Termination I was subdivided into four intervals based on different changing rates in the CO2 record (Monnin
et al., 2001). Depending on the realisation of the variation of the terrestrial carbon storage over time, three different
scenarios are discussed (A-TB0, A-TB1, A-TB2; Fig. 7). In the Keeling plot of these combined results the only
well distinct feature found in all three scenarios is the sharp drop in δ 13 C in parallel with rising pCO2 which occurs
during interval I (Fig. 15B). In the Taylor Dome data set this is also the only event in which the signal-to-noiseratio is high enough to allow a distinct identification. Changes in interval II and III occur during rather small or
no variations in pCO2 leading to features which are rather indistinguishable in the light of their uncertainties in
the data sets. The scenario A-TB2 shows a variability in δ 13 C which is closest to that of the Taylor Dome record,
while pCO2 is best reconstructed in the A-TB0 simulation.
In the following we have a closer look on the Keeling plots of these three combined scenarios and especially on
the results of A-TB2 (Fig. 15B, 16). Regression functions are applied to seven time windows in scenario A-TB2
(Table 4, Fig. 16B):
1. The main dynamics of a reduced marine export production due to iron limitation falls in time window 1
(20.0 − 17.1 kyr BP). In the combined scenario y0 is −7.3h while the decrease in export production as
single process would lead to y0 = −8.6h with at maximum −9.7h during the first years of the reduced
export production.
P. Köhler, J. Schmitt, H. Fischer
II
III IV
H1 BA YD
pCO2 [ atm]
A-TB0
A-TB1
A-TB2
data
280
260
240
220
200
180
20
-6.2
-6.2
-6.4
-6.4
-6.6
o
I
C [ /oo]
Interval
-6.6
-6.8
13
158
-6.8
-7.0
270
pCO2 [ atm]
230 210 190
170
B
-7.0
0.004
0.005
1/pCO2 [1/ atm]
A
18
16 14 12
Time [kyr BP]
10
Figure 15: A: Measured and simulated atmospheric carbon records over Termination I: Data: Atmospheric δ 13 C in
the Taylor Dome ice core (Smith et al., 1999); pCO2 in the EPICA Dome C ice core divided in four intervals with
different changing rates (Monnin et al., 2001). Both data set are synchronized to the GISP2 age scale (Meese et al.,
1997). The intervals II, III, and IV are approximately identical with the Heinrich 1 event (H1), the Bølling-Allerød
warm interval (BA), and the Younger Dryas cold event (YD) in the North Atlantic region. Simulation scenarios
combine all physical processes (ocean temperature, sea level, sea ice) with changes in ocean circulation (NADW
formation, Southern Ocean mixing), marine export production, CaCO3 compensation and terrestrial biosphere.
Differences in the scenarios A-TB0, A-TB1, and A-TB2 are the simulated changes the terrestrial carbon storage,
which are shown in Fig. 7. B: Keeling plot. Note, that the measured CO2 data are given in volume mixing ratio
[ppmv], while the model calculates the partial pressure (pCO2 ) in units of µatm.
2. The increase in Southern Ocean vertical mixing is mainly happening in window 2 (17.0 − 16.9 kyr BP)
leading to y0 = −10.0h in A-TB2, but to −8.2h as single process.
3. There is no specific process in operation in window 3 (16.8 − 16.6 kyr BP) which has a y0 of −5.2h.
4. In window 4 (16.5 − 14.9 kyr BP) covering the Heinrich 1 event the NADW formation is shut-off. Additionally, the marine export production is reduced by ∼10% due to macro-nutrient limitation.
5. NADW formation is resuming to intermediate strength in time window 5 covering the beginning of the
Bølling-Allerød warm interval in-between Heinrich 1 and the Younger Dryas (14.8 − 14.3 kyr BP). From
a comparison of the three scenarios (Fig. 15B) we further know that the changing carbon storage in the
terrestrial biosphere is additionally affecting the time windows 4 and 6, and that it is mainly responsible
for the dynamics in time window 5. In the scenarios A-TB0 and A-TB1 the time windows 4 and 5 are not
distinguishable.
6. In the Younger Dryas (window 6, 14.2 − 11.8 kyr BP) the NADW formation is partly reduced.
7. No clear regression (r2 = 6%) is found for window 7 (11.7 − 10.0 kyr BP), in which the carbon cycle is
organised to a new interglacial equilibrium.
A-TB2 leads to y-axis intercepts of −11.4, −13.3, and −13.0h in the windows 4, 5, and 6 respectively. For
comparison, the single process analysis has found y0 = −5.3 to −7.2h for changes in NADW formation between
16.5 and 11.8 kyr BP, but much higher negative values (−39.2h) in certain times. The rise in the terrestrial carbon
storage has y0 = −8.6 to −9.1h.
This analysis above shows that it seems possible to identify single processes in the Keeling plot of our artificial
time series, if they dominate the atmospheric carbon records for a certain time. However, due to the variety of
the processes the y-axis intercept which we identify in the combined scenarios is in no case near or close to the
expected y0 values we concluded from our single process analysis.
On the application and interpretation of Keeling plots in paleo climate research
-6.2
-6.4
-6.4
240
o
220
C [ /oo]
pCO2
13
C
-6.6
pCO2 [ atm]
230
190 170
o
2
1: y0=-07.3 /oo, r = 65%
o
2
2: y0=-10.0 /oo, r =100%
o
2
3: y0=-05.2 /oo, r = 97%
o
2
4: y0=-11.4 /oo, r = 83%
o
2
5: y0=-13.3 /oo, r = 90%
o
2
6: y0=-13.0 /oo, r = 76%
o
2
7: y0=-06.9 /oo, r = 06%
-6.6
13
pCO2 [ atm]
260
270
-6.2
159
200
180
-6.8
-6.8
-7.0
10
-7.0
A
160
20
18
16 14 12
Time [kyr BP]
B
0.004
0.005
0.006
1/pCO2 [1/ atm]
Figure 16: A detailed look at scenario A-TB2 (colors) in comparision to the Taylor Dome δ 13 C and EPCIA Dome
C CO2 data (grey) during Termination I (10 − 20 kyr BP). A: pCO2 (open circles), δ 13 C (closed circles); B:
Keeling plot. Seven time windows showing different dynamics and their regression functions are marked with
different colors. Note, that the measured CO2 data are given in volume mixing ratio [ppmv], while the model
calculates the partial pressure (pCO2 ) in units of µatm.
6 Discussion and Conclusions
In this study we analysed processes which alter the atmospheric content of carbon dioxide and δ 13 C of CO2 using
artificial time series produced with a global carbon cycle box model. Although there is evidence that the scenarios
investigated here are plausible and they can explain the observations during the last glacial/interglacial transition,
alternatives can not be ruled out due to data uncertainties and model simplicity. Nevertheless, for the investigation
of these artificial time series and their potential to be interpreted using the Keeling plot approach the absolute
validity of these scenarios is not important. By using a simple carbon cycle model we benefit from the fact that
individual processes acting on the carbon cycle can be switched on and off and their hypothetical impacts can be
analysed individually.
All processes have been analysed with respect to their impact to the Keeling plot analysis. A summary is found
in Table 2. The effective isotopic signature δ ∆A of terrestrial carbon uptake or release modifying the global carbon
cycle can be understood based on theoretical considerations of a three reservoir system which also includes oceanic
carbon uptake. These considerations can be understood as the paleo extension of the Keeling plot approach. The
∆A
effective isotopic signature converts to a boundary δδC→0
of −8.4h for terrestrial carbon fluxes approaching
∆A
zero. The δδC→0 obtained from theory is comparable to the y-axis intercept y0 in a classical Keeling plot. We
identified y-axis intercepts of different processes and compared it with the theoretical well understood value of the
Table 4: Regression analysis of time windows in the combined scenario A-TB2 covering carbon cycle dynamics
during Termination I. Results in 100 years resolution.
#
Time (kyr BP)
1
2
3
4
5
6
7
20.0 − 17.1
17.0 − 16.9
16.8 − 16.6
16.5 − 14.9
14.8 − 14.3
14.2 − 11.8
11.7 − 10.0
y0 (h)
r 2 (%) in brackets
−7.3 (65)
−10.0 (100)
−5.2 (97)
−11.4 (83)
−13.3 (90)
−13.0 (76)
−6.9 (6)
Main processes
Reduction in marine export production
Increase in Southern Ocean mixing
nothing special
NADW and export production reduced in Heinrich event 1 and terrestrial carbon storage
NADW rise in Bølling/Allerød and mainly terrestrial carbon storage
NADW reduced in Younger Dryas and and terrestrial carbon storage
nothing special, equilibration to interglacial climate
160
P. Köhler, J. Schmitt, H. Fischer
terrestrial carbon release. The y0 ’s of the prior/after analysis of our single process analysis vary between −0.7
and −8.6h. Some processes are compounds of subprocesses which have distinctively different y0 values (e.g. sea
ice cover in different hemispheres, y0 (north) = −4.8h; y0 (south) = −77.2h), in others the prior/after analysis
leads to very different results if a specific narrow time window is observed (e.g. terrestrial carbon storage, NADW
formation). Furthermore, the terrestrial carbon storage very likely changed in a non-linear way during the last
glacial/interglacial transition and not in a linear way as assumed in scenario TB0 (Fig. 7). This is also supported
by a simulation studies using a dynamical global simulation model (Köhler et al., 2005b). High frequency changes
on a centennial to millennial time scale are very likely smoothed out in the ice core records due to the average
mixing time of the air of several centuries and the limited sampling frequency. The single process analysis of
the states prior and after the experiments leads to y0 different than the boundary caused by terrestrial processes.
∆A
They vary from δδC→0
by a small decline (−0.2h) to a large increase (+7.7h). Processes in which the biology
∆A
is involved (marine and terrestrial) have identical y0 ’s (small decline in comparison to δδC→0
) and are therefore
∆A
indistinguishable. In all other cases y0 is more positive than δδC→0 .
The Taylor Dome record of atmospheric δ 13 C shows dynamics, in which a Keeling plot analysis leads to y0
values around −9.5h (Smith et al., 1999; Fischer et al., 2003). This holds, however only for the Holocene and the
climatically relatively stable LGM, while during the transition no clear end member could be recognised so far.
Unpublished measurements (Eyer, 2004) performed at the EPICA Dome C ice core show evidence for millennial
scale variability with fast and large changes in δ 13 C (more than 0.5h in a century). We believe from our analysis
that the processes responsible for these variations can not be identified based on a Keeling plot analysis. However,
our study gives us some indications which processes are in general able to contribute to fast changes and thus might
be responsible for very negative y-axis intercepts in a Keeling plot analysis. From what we have learned from our
model of the global carbon cycle fast changes in terrestrial carbon storage, variations in ocean circulation strength
(NADW formation) and sea ice coverage in the Southern Ocean are the only processes which can contribute
significantly to these δ 13 C excursions. However, the signal of fast terrestrial carbon release is smoothed by air
mixing processes in the firn. In the EPICA Dome C ice core with its small accumulation rate and large mixing
time terrestrial processes might therefore not be responsible for these negative δ 13 C signals.
In the light of these conclusions we have to acknowledge that especially the modelling response to Southern
Ocean sea ice extent is highly model dependent. Large variability in pCO2 and δ 13 C occur especially when the
surface ocean box is nearly fully covered with sea ice and thus reducing gas exchange dramatically (Köhler and
Fischer, 2006). Thus, our model architecture with a Southern Ocean box ranging from 40◦ S to the Antarctic continent is probably too simplistic to cover the complete picture of observed dynamics. Other model intercomparisions
have already pointed out these model-dependent behaviour (Archer et al., 2003).
From the understanding which emerges here, it seems unlikely, that the interpretation of δ 13 C measured during
glacial/interglacial transitions can be enhanced very much with the Keeling plot approach. The identification of
a single process which might be responsible for the observed fluctuations in atmospheric CO2 and δ 13 C can not
be based on a Keeling plot analysis. Most processes acted simultaneously on the global carbon cycle during the
transition and the uncertainties in data retrieval, y-axis intercept, and δ 13 C are too large to come to a sound and
unequivocal process identification.
Acknowledgements
We thank J. Severinghaus for ideas on the three box model calculations and J. Freitag for discussions on firnification and bubble close off. This study is funded by the German Ministry of Education and Research through the
German Climate Research Programme DEKLIM (project RESPIC).
References
Archer, D. E., Martin, P. A., Milovich, J., Brovkin, V., Plattner, G.-K., and Ashendel, C.: Model sensitivity
in the effect of Antarctic sea ice and stratification on atmospheric pCO2 , Paleoceanography, 18, 1012, doi:
10.1029/2002PA000 760, 2003.
Blasing, T. J., Broniak, C., and Marland, G.: Estimates of monthly carbon dioxide emissions and associated δ 13 C
values from fossil-fuel consumption in the U.S.A, 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, 2004.
On the application and interpretation of Keeling plots in paleo climate research
161
Bowling, D. R., Tans, P. P., and Monson, R. K.: Partitioning net ecosystem carbon exchange with isotopic fluxes
of CO2 , Global Change Biology, 7, 127–145, 2001.
Brook, E. J., Harder, S., Serveringhaus, J., Steig, E. J., and Sucher, C. M.: On the origin and timing of rapid
changes in atmospheric methane during the last glacial period, Global Biogeochemical Cycles, 14, 559–572,
2000.
Crosta, X., Pichon, J.-J., and Burckle, L. H.: Application of modern analog technique to marine Antarctic diatoms:
Reconstruction of maximum sea-ice extent at the Last Glacial Maximum, Paleoceanography, 13, 284–297,
1998a.
Crosta, X., Pichon, J.-J., and Burckle, L. H.: Reappraisal of Antarctic seasonal sea-ice extent at the Last Glacial
Maximum, Geophysical Research Letters, 14, 2703–2706, 1998b.
Dansgaard, W., Clausen, H. B., Gundestrup, N., Hammer, C. U., Johnsen, S. F., Kristinsdottir, P. M., and Reeh, N.:
A new Greenland deep ice core, Science, 218, 1273–1277, 1982.
Emanuel, W. R., Killough, G. G., Post, W. M., and Shugart, H. H.: Modeling terrestrial ecosystems in the global
carbon cycle with shifts in carbon storage capacity by land-use change, Ecology, 65, 970–983, 1984.
Etheridge, D. M., Steele, L. P., Langenfelds, R. L., Francey, R. J., Barnola, J.-M., and Morgan, V. I.: 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.
Eyer, M.: Highly resolved δ 13 C measurements on CO2 in air from Antarctic ice cores, Ph.D. thesis, University of
Bern, Bern, Switzerland, 2004.
Eyer, M., Leuenberger, M., Nyfeler, P., and Stocker, T.: Comparison of two δ 13 CO2 records measured on air from
the EPICA Dome C and Kohnen Station ice cores, Geophysical Research Abstracts, 6, 01 990, 2004.
Fairbanks, R. G.: The age and origin of the Younger Dryas climate event in Greenland ice cores, Paleoceanography,
5, 937–948, 1990.
Fischer, H., Wahlen, M., and Smith, J.: Reconstruction of glacial/interglacial changes in the global carbon cycle
from CO2 and δ 13 CO2 in Antarctic ice cores, Memoirs of the National Institute for Polar Research, Special
Issue, 57, 121–138, 2003.
Flanagan, L. B. and Ehleringer, J. R.: Ecosystem-atmosphere CO2 exchange: interpretating signals of change
using stable isotope ratios, Trends in Ecology and Evolution, 13, 10–14, 1998.
Francey, R. J., Allison, C. E., Etheridge, D. M., Trudinger, C. M., Enting, I. G., Leuenberger, M., Langenfelds,
R. L., Michel, E., and Steele, L. P.: A 1000-year high precision record of δ 13 C in atmospheric CO2 , Tellus, 51B,
170–193, 1999.
Friedli, H., Lötscher, H., Oeschger, H., Siegenthaler, U., and Stauffer, B.: Ice core record of the 13 C/12 C ratio of
atmospheric CO2 in the past two centuries, Nature, 324, 237–238, 1986.
Gersonde, R., Crosta, X., Abelmann, A., and Armand, L.: Sea-surface temperature and sea ice distribution of
the Southern Ocean at the EPILOG Last Glacial Maximum — a circum-Antarctic view based on siliceous
microfossil records, Quaternary Science Reviews, 24, 869–896, 2005.
Hemming, D., Yakir, D., Ambus, P., Aurela, M., Besson, C., Black, K., Buchmann, N., Burlett, R., Cescatti,
A., Clement, R., Gross, P., Granier, A., Grünwald, T., Havrankova, K., Janous, D., Janssens, I. A., Knohl,
A., Köstner, B., Kowalski, A., Laurila, T., Mata, C., Marcolla, B., Matteucci, G., Moncrieff, J., Moors, E. J.,
Osborne, B., Pereira, J. S., Pihlatie, M., Pilegaard, K., Ponti, F., Rosova, U., Rossi, F., Scartazza, A., and Vesala,
T.: Pan-European δ 13 C values of air and organic matter from forest ecosystems, Global Change Biology, 11,
1065–1093, doi: 10.1111/j.1365–2486.2005.00 971.x, 2005.
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.
Johnsen, S. J., Clausen, H. B., Dansgaard, W., Fuhrer, K., Gundestrup, N., Hammer, C. U., Iversen, P., Jouzel, J.,
Stauffer, B., and Steffensen, J. P.: Irregular glacial interstadials recorded in a new Greenland ice core, Nature,
359, 311–313, 1992.
162
P. Köhler, J. Schmitt, H. Fischer
Keeling, C. D.: The concentration and isotopic abundance of carbon dioxide in rural areas, Geochimica et Cosmochimica Acta, 13, 322–334, 1958.
Keeling, C. D.: The concentration and isotopic abundance of carbon dioxide in rural and marine air, Geochimica
et Cosmochimica Acta, 24, 277–298, 1961.
Keeling, C. D. and Whorf, T. P.: 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., USA, 2005.
Keeling, C. D., Bollenbacher, A. F., and Whorf, T. P.: Monthly atmospheric 13 C/12 C isotopic ratios for 10 SIO
stations, 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.
Köhler, P. and Fischer, H.: Simulating changes in the terrestrial biosphere during the last glacial/interglacial transition, Global and Planetary Change, 43, 33–55, doi: 10.1016/j.gloplacha.2004.02.005, 2004.
Köhler, P. and Fischer, H.: Proposing a mechanistic understanding of changes in atmospheric CO2 during the last
740 000 years, Climate of the Past Discussions, 2, 1–42, SRef–ID: 1814–9359/cpd/2006–2–1, 2006.
Köhler, P., Fischer, H., Munhoven, G., and Zeebe, R. E.: Quantitative interpretation of atmospheric carbon records
over the last glacial termination, Global Biogeochemical Cycles, 19, GB4020, doi: 10.1029/2004GB002 345,
2005a.
Köhler, P., Joos, F., Gerber, S., and Knutti, R.: Simulated changes in vegetation distribution, land carbon storage, and atmospheric CO2 in response to a collapse of the North Atlantic thermohaline circulation, Climate
Dynamics, 25, 689–708, doi: 10.1007/s00 382–005–0058–8, 2005b.
Köhler, P., Muscheler, R., and Fischer, H.: A model-based interpretation of low frequency changes in the carbon
cycle during the last 120 kyr and its implications for the reconstruction of atmospheric ∆14 C and the 14 C production rates estimates, Geochemistry, Geophysics, Geosystems, p. submitted; doi: 10.1029/2005GC001228,
2006.
Landais, A., Barnola, J. M., Masson-Delmotte, V., Jouzel, J., Chappellaz, J., Caillon, N., Huber, C., Leuenberger, M., and Johnsen, S. J.: A continuous record of temperature evolution over a sequence of DansgaardOeschger events during Marine Isotope Stage 4 (76 to 62 kyr BP), Geophysical Research Letters, 31, L22 211,
doi: 10.1029/2004GL021 193, 2004.
Lang, C., Leuenberger, M., Schwander, J., and Johnsen, S.: 16◦ C rapid temperature variation in central Greenland
70,000 years ago, Science, 286, 934–937, 1999.
Levin, I., Ciais, P., Langenfelds, R., Schmidt, M., Ramonet, M., Sidorov, K., Tchebakova, N., Gloor, M., Heimann,
M., Schulze, E.-D., Vygodskaya, N. N., Shibistova, O., and Lloyd, J.: Three years of trace gas observations over
the EuroSiberian domain derived from aircraft sampling – a concerted action, Tellus, 54B, 696–712, 2002.
Marland, G., Boden, T., and 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.
Meese, D. A., Gow, A., Alley, R., Zielinski, G., Grootes, P., Ram, M., Taylor, K., Mayewski, P., and Bolzan, J.:
The Greenland Ice Sheet Project 2 depth-age scale: Methods and results, Journal of Geophysical Research, 102,
26 411–26 423, 1997.
Monnin, E., Indermühle, A., Dällenbach, A., Flückiger, J., Stauffer, B., Stocker, T. F., Raynaud, D., and Barnola,
J.-M.: Atmospheric CO2 concentrations over the last glacial termination, Science, 291, 112–114, 2001.
Mook, W. G.:
13
C in atmospheric CO2 , Netherlands Journal of Sea Research, 20, 211–223, 1986.
Munhoven, G.: Modelling glacial-interglacial atmospheric CO2 variations: the role of continental weathering,
Ph.D. thesis, Université de Liège, Liège, Belgium, 1997.
Pataki, D. E., Ehleringer, J. R., Flanagan, L. B., Yakir, D., Bowling, D. R., Still, C. J., Buchmann, N., Kaplan,
J. O., and Berry, J. A.: The application and interpretation of Keeling plots in terrestrial carbon cycle research,
Global Biogeochemical Cycles, 17, 1022, doi: 10.1029/2001GB001 850, 2003.
On the application and interpretation of Keeling plots in paleo climate research
163
Petit, J. R., Jouzel, J., Raynaud, D., Barkov, N. I., Barnola, J.-M., Basile, I., Bender, M., Chappellaz, J., Davis, M.,
Delaygue, G., Delmotte, M., Kotlyakov, V. M., Legrand, M., Lipenkov, V. Y., Lorius, C., Pépin, L., Ritz, C.,
Saltzman, E., and Stievenard, M.: Climate and atmospheric history of the past 420,000 years from the Vostok
ice core, Antarctica, Nature, 399, 429–436, 1999.
Plattner, G.-K., Joos, F., and Stocker, T. F.: Revision of the global carbon budget due to changing air-sea oxygen
fluxes, Global Biogeochemical Cycles, 16, 1096, doi: 10.1029/2001GB001 746, 2002.
Ridgwell, A. J.: Glacial-interglacial perturbations in the global carbon cycle, Ph.D. thesis, University of East
Anglia, Norwich, U.K., 2001.
Sabine, C. L., Feely, R. A., Gruber, N., Key, R. M., Lee, K., Bullister, J. L., Wanninkhof, R., Wong, C. S., Wallace,
D. W. R., Tilbrook, B., Millero, F. J., Peng, T.-H., Kozyr, A., Ono, T., and Rios, A. F.: The oceanic sink for
anthropogenic CO2 , Science, 305, 367–371, 2004.
Sarnthein, M., Pflaumann, U., and Weinelt, M.: Past extent of sea ice in the northern North Atlantic inferred from
foraminiferal paleotemperature estimates, Paleoceanography, 18, 1047, doi: 10.1029/2002PA000 771, 2003.
Scholze, M., Kaplan, J. O., Knorr, W., and Heimann, M.: Climate and interannual variability of the atmospherebiosphere 13 CO2 flux, Geophysical Research Letters, 30, 1097, doi: 10.1029/2002GL015 631, 2003.
Schwander, J. and Stauffer, B.: Age difference between polar ice and the air trapped in its bubbles, Nature, 311,
45–47, 1984.
Schwander, J., Jouzel, J., Hammer, C. U., Petit, J.-R., Udisti, R., and Wolff, E.: A tentative chronology for the
EPICA Dome Concordia ice core, Geophysical Research Letters, 28, 4243–4246, 2001.
Siegenthaler, U., Stocker, T. F., Monnin, E., Lüthi, D., Schwander, J., Stauffer, B., Raynaud, D., Barnola, J.M., Fischer, H., Masson-Delmotte, V., and Jouzel, J.: Stable carbon cycle-climate relationship during the late
Pleistocene, Science, 310, 1313–1317; doi: 10.1126/science.1120 130, 2005.
Smith, H. J., Fischer, H., Wahlen, M., Mastroianni, D., and Deck, B.: Dual modes of the carbon cycle since the
Last Glacial Maximum, Nature, 400, 248–250, 1999.
Steig, E. J., Brook, E. J., White, J. W. C., Sucher, C. M., Bender, M. L., Lehman, S. J., Morse, D. L., Waddington,
E. D., and Clow, G. D.: Synchronous climate change in Antarctica and North Atlantic, Science, 282, 92–95,
1998a.
Steig, E. J., Morse, D. L., Waddington, E. D., and Polissar, P. J.: Using the sunspot cycle to date ice cores,
Geophysical Research Letters, 25, 163–166, 1998b.
Sturm, P., Leuenberger, M., and Schmidt, M.: Atmospheric O2 , CO2 and δ 13 C observations from the remote
sites Jungfraujoch, Switzerland, and Puy de Dôme, France, Geophysical Research Letters, 32, L17 811; doi:
10.1029/2005GL023 304, 2005.
Trudinger, C. M., Enting, I. G., Francey, R. J., Etheridge, D. M., and Rayner, P. J.: Long-term variability in the
global carbon cycle inferred from a high-precision CO2 and δ 13 C ice-core record, Tellus, 51B, 233–248, 1999.
Volk, T. and Hoffert, M. I.: Ocean carbon pumps: analysis of relative strengths and efficiencies in ocean-driven
atmospheric CO2 changes, in The carbon cycle and atmospheric CO2 : Natural variations archean and present,
edited by E. T. Sundquist and W. S. Broecker, vol. 32 of Geophysical Monograph, pp. 99–110, American
Geophysical Union, Washington, D.C., USA, 1985.
Wolff, E. W., Kull, C., Chappellaz, J., Fischer, H., Miller, H., Stocker, T. F., Watson, A. J., Flower, B., Joos,
F., Köhler, P., Matsumoto, K., Monnin, E., Mudelsee, M., Paillard, D., and Shackleton, N.: Modeling past
atmospheric CO2 : results of a challenge, EOS, 86 (38), 341, 345, 2005.
Yakir, D. and Sternberg, L. D. S. L.: The use of stabile isotopes to study ecosystem gas exchange, Oecologia, 123,
297–311, 2000.
Zeebe, R. E. and Wolf-Gladrow, D. A.: CO2 in Seawater: Equilibrium, Kinetics, Isotopes, vol. 65 of Elsevier
Oceanography Book Series, Elsevier Science Publishing, Amsterdam, The Netherlands, 2001.
164
165
Acknowledgements
My first thank goes to Prof. Heinz Miller, whose long-term effort to bring together an enthusiastic team of specialists from various disciplines to cover the entire deep ice core business,
beginning with the reconnaissance of a proper drill site, the deep drilling, a broad spectrum of
analytical approaches for physical and chemical aspects of the ice, and towards a climatic
interpretation of the these data sets including modeling, made this PhD thesis possible. I thank
Prof. Kai-Uwe Hinrichs for being the ‘Zweitgutachter’ of this thesis.
I’m very indebted to Hubertus Fischer, for supporting me in any aspect of scientific life. He
had the scientific curiosity to discuss about all my newest theories, why this and that doesn’t
work and how it might work - no matter if he were in the field in Antarctica or in New York.
And thanks for your optimism, which helped me go further and further.
Many thanks to…..
Anna Wegner and Felix Fundel for being the best room mates and enriching my daily life,
especially during the desktop phase, with fun & jokes, cakes, matlab conversations, playing
boule at the beach, and for numerous Kefirs and orange juices to keep me healthy.
Melanie Behrens and Ulrike Salzer for their help in the mass spec lab and providing technical
support. And above all, thanks for working on similar problems. Klaus-Uwe Richter for his
nice help with questions concerning electronic equipment at the beginning. Jens Hefter for
collegially sharing the mass spec lab with me. Andreas Frenzel for selecting the right computer for me.
Peter Köhler for fruitful discussions about the global carbon cycle and the opportunity to see
how a modeler sees the world. Johannes Freitag for his help to get in touch with the world of
firnification and for all the brief philosophical excursions while walking from the ‘Café Caspar’ back to AWI.
Thanks to the entire glaciology group for the stimulating working atmosphere and especially
Hans Oerter. Further, Fernando Valero-Delgado for singing on the floor, taking care of all the
ice core boxes, for all the cakes he organized. Sepp Kipfstuhl for always keeping a critical eye
on the cold labs and his exceptional scientific spirit and curiosity.
166
Birthe Twarloh for continuously reminding me to fulfill my social obligations, to decide
Monday what to like Friday, and for the feeling to be a real scientist. All the colleagues, who
left the group: Wolfgang Rack, Urs Ruth, Olaf Eisen, Anja Lambrecht, Marie-Luise SiggardAndersen, for their special contribution to enrich life within the glaciology group. Michael
Bock for doubling the number of Geoecologists within the glaciology group.
The night porters, especially Heinz, for their nice words when suddenly approaching me during the night in one of my laboratories. The secretaries of the department and also the general
administration of AWI for their patience if I repeatedly forgot to sign a form.
The drilling and science team of the field campaign 2003/2004 at the Dronning Maud Land
drill site – it was a great experience to be in Antarctica and learn to drill ice cores, thanks especially to Frank Wilhelms. The EPICA community, especially all the nice colleagues from
Bern, who shared data and ideas and for the good scientific spirit and social life on all the
conferences I could attend.
Sonja Paul for her love and the hundred things which makes life worth living and which
helped to forget the isotopes for while. A special thank goes to the beautiful beech forests and
the landscape in the surroundings of Göttingen, to which I could escape and who provided me
with everything I missed in Bremerhaven.
Finally, I am indebted to my parents and family for their support during my education all the
years of my studies and for being the lovely home to which I came back too rarely.
167
Erklärung
Ich versichere, dass ich die vorliegende Dissertation selbständig verfasst und keine anderen
als die angegebenen Quellen und Hilfsmittel verwendet habe.
Ich versichere, dass ich nicht bereits anderweitig eine Dissertation eingereicht oder versucht
habe, mich einer Doktorprüfung zu unterziehen.
Bremerhaven, den 02.10.2006
________________________
(Jochen Schmitt)
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