!"#$%"&'%($#)&*+),-&%).&/01+(&2%(+%$+"+34&5)61((1.&6(#7&837#-901(+:&;<&%).&=>?'
8@30#(A-BC&DE&F%33"1G&DE&HE&F1).1(G&IE&IE&/%)-G&JE&KE&'E&K0+31G&JE&/E&L""+-G&/E&'#)M%4G&NE&JE
O(%):14
*#@(:1C&*:+1):1G&P1M&*1(+1-G&2#"E&<QRG&P#E&STU<&AD%(E&?>G&<VVVBG&99E&<TURW<TRV
I@$"+-01.&$4C&871(+:%)&8--#:+%3+#)&6#(&301&8.X%):171)3&#6&*:+1):1
*3%$"1&YNHC&http://www.jstor.org/stable/3074760
8::1--1.C&?VZV?Z<VV[&>QCV<
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=aaas.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the
scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that
promotes the discovery and use of these resources. For more information about JSTOR, please contact [email protected].
American Association for the Advancement of Science is collaborating with JSTOR to digitize, preserve and
extend access to Science.
REPORTS
Global
Carbon
Variability
Sinks
Inferred
and
Their
from
Atmospheric 02 and 613C
M. Battle,'* M. L. Bender,l P. P. Tans,2 J. W. C. White,3
J. T. Ellis,4 T. Conway,2 R. J. Francey5
Recenttime-series measurementsof atmospheric02 show that the landbiosphere
and worldoceans annuallysequestered 1.4 ? 0.8 and 2.0 ? 0.6 gigatons of carbon,
respectively,between mid-1991 and mid-1997.The rapidstorage of carbonby the
land biosphere from 1991 to 1997 contrasts with the 1980s, when the land
biospherewas approximatelyneutral.Comparisonwith measurementsof 813CO2
impliesan isotopicflux of 89 + 21 gigatonsof carbonpermil peryear, in agreement
with model- and inventory-basedestimates of this flux. Boththe 813Cand the 02
data show significantinterannualvariabilityin carbonstorage over the periodof
record.The general agreement of the independentestimates from 02 and 13Cis
a robustsignalof variablecarbonuptakeby both the landbiosphereandthe oceans.
Between 1991 and 1997, combustionof fossil
fuels added roughly 6.2 gigatons of carbon
per year (GtC/year)(1, 2) to the atmosphere
in the form of CO2.During this same period,
the atmosphericburdenof CO2 increasedby
only 2.8 GtC/year. The balance of the CO2
was taken up by the oceans and the land
biosphere. Determiningthe ocean-landpartition of carbonuptake (and the temporalvariability of the partition) is essential for any
mechanisticunderstandingof carbonstorage.
We use two approachesto experimentally
constrain this partition. The first approach
involves measuringthe rate of change of the
02 concentrationof air (actually, the 02/N2
ratio). Fossil fuel combustion decreases atmospheric 02/N2, and terrestrialgrowth releases 02, attenuatingthe fossil fuel decrease
in this ratio. Ocean uptake of CO2 does not
affect 02/N2. Thus, the change in atmospheric 02/N2 indicates terrestrialcarbon storage
(3, 4). The second approach involves measuringthe change in the isotopic composition
(513C)of atmosphericCO2. Fossil fuels possess low 13Cvalues. Duringphotosynthesis,
terrestrial plants assimilate carbon with a
lower s13C value, therebyenrichingthe 63C
of the CO2 left behind in the atmosphere.
Moreover, net uptake of CO2 between the
'Department of Geoscience, Guyot Hall, Princeton
University, Princeton, NJ 08544 USA. ZNational Oceanic and Atmospheric Administration/Climate Monitoring and Diagnostics Laboratory, R/E/CG1, 325
Broadway, Boulder, CO 80303 USA. 3lnstitute for
Arctic and Alpine Research, and Department of
Geological Sciences, University of Colorado, Boulder, CO 80309, USA. 4Graduate School of Oceanography, University of Rhode Island, South Ferry
Road, Narragansett, RI 02882, USA. 5Division of
Atmospheric Research, Commonwealth Scientific
and Industrial Research Organization, Mordialloc,
Victoria 3195, Australia.
*Present address: Department of Physics and Astronomy, Bowdoin College, 8800 College Station, Brunswick, ME 04011, USA.
atmosphereand oceans leaves 513Cessentially unchanged. Thus, the change in atmospheric 613C gives a measure of terrestrial
carbon storage (5).
These approaches to determining the
partition of land and ocean uptake have
geochemical uncertainties that may be
large. Uncertainty in the 02 method arises
because the ocean may be a small net
source or sink of 02 over the course of a
year or more. Every spring, atmospheric
02/N2 rises by about 60 per meg [or 0.06
per mil (%o)](5) because of ocean biology.
Net production of organic matter releases
02, which supersaturates the mixed layer
and escapes to the atmosphere. Every winter, 02/N2 falls by 60 per meg as the oceans
take up oxygen to resaturatesuboxic waters
of the interior (6-8). This seasonal cycle is
-15 times larger than the annual change in
atmospheric 02/N2 because of net carbon
storage by the land biosphere. These natural oceanic fluxes may not be in exact
balance over the course of 1 year (or even
several). A natural imbalance in these fluxes would cause a change in the annual
average 02/N2 ratio of air. We would erroneously attributethis change to carbon storage by the land biosphere (4, 7, 8). Such an
imbalance is unlikely to persist for more
than a few years. Thus, carbon fluxes based
on 02 are most robust when averaged over
many years.
Uncertainty in the 613C method comes
from uncertainties in isofluxes (also called
"gross" or "disequilibrium"fluxes). Every
year, large quantitiesof carbonpass through
the land biosphere.Any net storageor release
is an imbalance between two large fluxes:
-50 Gt of carbon fixed by photosynthesis
and -50 GtC released through respiration.
The g63C of assimilated CO2 reflects the
modem atmospheric s13C, whereas the 813C
of respired CO2 reflects the "13Cof the at-
www.sciencemag.org SCIENCEVOL287
mosphere of years past, when the plant assimilated the carbon that is now being respired. Similarly,though the oceans annually
sequesterroughly 2 Gt of carbon,the oceans
and atmosphere exchange about 90 GtC.
Consequently,the flux of CO2 into the ocean
carries the current signature of atmospheric
813C, whereas the flux of CO2 out of the
ocean carries an atmospheric813C signature
that is several decades old. In both cases, the
isofluxes greatly attenuatethe change in atmospheric 13C that would otherwise occur.
Estimates of the isofluxes reflect uncertainties in the age spectrumof decomposing organic matter, decomposition rates as a function of latitude, switches from C3 to C4
photosynthesis,variabilityin the 613Cof CO2
in the upper ocean, the isotopic composition
of fossil fuel, and rates of gas exchange at the
air-sea interface (9, 10). Although these uncertainties are substantial, they arise from
processes that, when globally averaged, are
largely constant from one year to the next
(with the possible exception of switches between C3 and C4 photosynthesis). Thus, interannualvariability in carbon fluxes is capturedby 613C, even if the long-termpartition
is ill-constrained.Moreover,the uncertainties
described here for '13C are essentially independent of those for 02/N2. Thus, information from one species can be used to reduce
uncertaintiesin the other.
Beginning in 1991, global 02/N2 measurement programs began (4, 8), and 613C
sampling networks and interlaboratorycomparisons expandeddramatically(11, 12). Results from these tracersfor the years 1991-94
(8, 13-15) showed that the land biosphere
was an importantCO2 sink. Here we present
updatedrecordsof both 813C and 02/N2 from
a global network of stations. With nearly 7
years of data in hand for both species, we are
able to assess the carbon balance through
1997 with more confidence than was previously possible, constrainthe isofluxes of carbon, and examine interannualvariability in
carbonstorage.Takenalong with otherrecent
02/N2-based analyses, these data indicate a
persistent shift in land biosphere sequestration rates as compared with those of the
previous decade.
We begin by calculating oceanic and terrestrial carbon fluxes using 02. The atmospheric mass balance for CO2 can be written
d(C02)
dt
-0.471
where
X (fiuel + ce.ent +fland
ffuel, fcement'
land'
+focen)
andfcean,
(1)
are flux-
es of carbon from the atmosphere due to
fossil fuel combustion, cement manufacturing, the terrestrialbiosphere, and the ocean,
respectively (all negative for release to the
atmosphere;focean and fland positive for se-
31 MARCH2000
2467
REPORTS
questration). The fluxes are in GtC/year,
and the atmospheric inventory term
d(CO2)/dt is in parts per million (ppm) per
year (2). Similarly, the mass balance for
atmospheric oxygen can be written
d(O2/N2)
These equations can be solved to give fland
and fcean in terms of measured quantities
1
1.43
d(O2/N2)
fland-
+
1.1 ffuel24
d / CO2
O2/N2)
Jocean
dt 0.471
2.49
dt
dt
(3)
1.1 - 1.43
4.8 X 0.471 X (1.43ff,e, + 1.l iand)
(2)
The coefficients offfel andland arise because
1.43 mol of 02 are consumed for every mole
of CO2 combusted for today's mix of fuels
(16), and 1.1 mol of 02 are produced for
every mol of CO2 sequestered by the land
biosphere (17). 0.471 converts GtC to ppm
CO2, and 4.8 converts ppm to per meg (5).
1.1
ffuel
fcement
(4)
Figure 1 shows measurements of O2/N2 and
CO2from our laboratories(18) as well as previously publisheddataof R. Keeling (15). We
use the fuel and cementproductionrecordsof
Marlandet al. (1). We assumethatfossil carbon
emissions increase in 1996 and 1997 at the
same rate as in 1995. We assume that at-
0-
390
-40-
380
-80 '
370
-120360
-160-200-
350
-240-
340
mospheric nitrogen is constant (16) and
that over the period of study, oceanic 02
fluxes are balanced.
Averaging the records from the Barrow/
Alert and Cape Grim stations, we calculate a
global ocean uptake of 2.0 ? 0.6 GtC/year
for the years 1991.5-1997.5 (see Table 1 for
errors)(19). As a check, we treatthe records
at Samoa (1993.5-1997.5) and Syowa
(1994.0-1997.0) as globally representative.
From these, we calculate ocean uptakeof 2.2
and 1.4 GtC/year.Unfortunately,these latter
records are too short for robust fits and error
estimates.
Based on the Cape Grimand Barrowdata,
we calculate a global average terrestrialcarbon sink (fland) of 1.4 ? 0.8 GtC/year.The
Samoa and Syowa recordsimply sinks of 0.9
and 1.4 GtC/year,respectively. These results
are shown in Fig. 3.
Having calculatedcarbonfluxes using 02/
N2, we use this information,along with measurementsof 813C,to constrainthe isofluxes.
Following the presentationof Tans et al. (20),
we can express the change in the atmospheric
abundanceof the '3C isotope as
d
CO2 d {5
133t}813
ffuelfuel
40U-
0
- 380
-80-
E
O
- 350 5O
-200-
CD -240-
40-
0
0;
-40-
-
340
cement -
1
atm}
/t-x
(Z)
is the kinetic fractionation that occurs
thesis by the land biosphere. fuel, fcement'
and fland all represent net (one-way)
foceand
fluxes. CO2is the total atmosphericinventory
in GtC. G captures the isoflux from gross
- 390
(two-way)
- 380
exchanges
between
the atmo-
-160-
- 360
sphere and the land and ocean reservoirs.
Values of e and 8 are in per mil. G has units
of GtC %o/year.Having determinedfocean
and
and CO2 measurements,
fland with the 02/N2
-200-
- 350
we can now solve for G.
-240-
_ 340
-80-120-
-40 -80 -120 -160 -240
1991.0
-
390
-
380
Valuesused in this calculationareshownin
Table 1. The 813Cdata (Fig. 2) are from the
NationalOceanicandAtmosphericAdministration-ClimateMonitoringandDiagnosticsLabfor Arctic
oratory/Colorado
University/Institute
_
370
and Alpine
-
-200 1992.0
1993.0
1994.0
1995.0
1996.0
1997.0
360
350
340
1998.0
Date
andCO2recordsmeasured(18) in aircollectedat six sites (335).Axeshave been
Fig. 1. The O02/N2
scaledso that removalsof 1 mol each of 02 and CO2resultin the same displac:ement.Thesecular
trend in 802/N2 is primarilydue to combustionof fossil fuel. The seasonalcyc:leof 802/N2 has a
terrestrialcomponentthat is the inverseof the CO2variationtimes the phot:osyntheticstoichiometry (1.1) (17). The oceanic contributionto the seasonal 6O2/N2 variationreflects photosynthesis and ventilationas discussedin the text. Thereis also a thermalcomponent to the cycle
drivenby changesin the sea surfacetemperatureand the differentsolubilities
sof 0 and N2 (7).
Alsoshownare adjustedvalues(18) of previouslypublishedSIOdata of Keelinget al. (15) for Alert
and CapeGrim(35) with respectto GF1[a flaskof standardair of arbitrarycomposition(8)].
2468
Jcement
during CO2 invasion into the ocean. Eair-land
representsthe fractionationduringphotosyn-
- 360
-160:
0.
air
- 370
-120Q
4)
-
p
_
, _
--w
- JoceanEair-sea- JlandEair-landt (J
- 390
0-40 -
atm}
31 MARCH2000
VOL287
Research (NOAA-CMDL/CU-
INSTAAR) measurementprogram(11). The
CO2 dataare fromthe NOAA-CMDLcooperative flask samplingnetwork(21). We findthat
G is 89 + 21 GtC %o/year.The errorcorrespondsto an uncertaintyin the land-oceanpartition of ? 1.3 GtC/year.This resultis in good
agreementwith the value of 83 GtC %o/year
derived using the method of Ciais et al. (9)
[updatedwith the observed(atmospheric)and
estimated(oceanic)evolutionof 813C].
In additionto the new results from this
study, Fig. 3 shows the land-oceanpartition
derivedfrom archivedair samplesby Langenfelds et al. (22), fromthe fim-airstudyof Battle
SCIENCEwww.sciencemag.org
REPORTS
et al. (23), andfrompreviouslypublishedwork
of Keelinget al. (15). Takenas a compositetime
series, the data show a substantialshift in the
global carboncycle. From 1977 to 1990, the
landbiospherewas neithera sourcenora sinkof
carbon. In contrast,the first 7 years of this
decade show a significantterrestrialsink of
carbon.Thisbroadtemporalfeatureagreeswith
the synthesis-inversionresult of Rayneret al.
(24), in part because many of our data were
common to that study. The oceanic sink is in
oceancirgood agreementwithtracer-calibrated
culationmodels (1.7 to 2.8 GtC/year)(25).
As a step towardunderstandingthe causes
of this shift,we examineinterannualvariations
in carbonfluxes. As discussed above, we expect most of the uncertaintiesin 613C-based
fluxesto be constantovera 7-yearrecord.Inthe
case of 02, however, we have little reason a
priorito expectthe oceanicproduction-ventilation processes driving02 fluxes to be in balance over a yearor two. Thus,we wouldexpect
6s3C-basedresults to reflect interannualvariabilityin fluxes more faithfullythan O2-based
results.This simple pictureis potentiallycomplicated by shifts in terrestrialphotosynthesis
from C3 to C4 pathwaysand by changes in
discriminationby C3 plantsdue to droughtor
nutrientstress.
Figure4 shows the 2-year averagedtrends
(26) in the land and ocean carbon sinks in-
in the land-ocean partition (13). Based on
measurements of the same samples from
1992-1998 and on measurementsof common
calibrationstandardsin 1994, 1995, and 1997,
discrepanciesbetweenthe CommonwealthScientific and IndustrialResearch Organization
P53C measurement
(CSIRO) and INSTAAR
programshave been observed(11, 12, 27). In
particular,a calibrationdifference of 0.03 to
to a 1-yearflux of 1.2 to
0.04 %o(corresponding
1.6 GtC) developed in mid-1994. We provisionally adoptthe CSIROcalibrationhere and
use it to correctfor this discrepancy.The remainingstandarddeviationof annualmeandifferences between the labs is then 0.014 %o,
correspondingto an uncertaintyof +0.5 GtC/
year in the land-oceanpartition.The largest
differenceoccurs for 1995, for which the unthereis
certaintyis twice as large.Furthermore,
no discernibletrendin the intercalibration
between INSTAARand CSIRO(beyondthe offset in 1994).
Errors in
02/N2
also translate directly into
the land-oceanpartition.All the stationsshown
in Fig. 4 exhibit consistenttrends,which suggests minimal collection bias. Frequentsampling leads to small statisticalfittingerrors(the
statisticalerroronfocea is +0.4 GtC/yearfor a
12-monthperiod at Cape Grim or Barrow).
Nonetheless,transientcalibrationartifactsprob-
1>1
LL
a
(3
mE 11
co
w This study (613C) --
Keeling1996(02) -
Samoa (02) *** Syowa (0) |
This study (0O) Langenfelds1999 (02) -- Battle 1996 (Firn02)
c
r-
T
ferred from 813C and 802/N2 measurements.
Based on the 813C records in Fig. 4, it ap-
pears that the terrestrial carbon sink was
strongestin 1992-93 and 1996-97 and weaker in 1994-95. The ocean appears to have
been a vigorous sink before 1995, with declining intensity in 1995-96. Terrestrialcarbon storageseems more variablethanoceanic
storage. Furthermore,the similar phasing of
the 13C- and O2-derived swings in uptake
implies that the oceanic 02 cycle is roughly
balanced over a 2-year period and that the
813C record is not solely reflecting switches
from C3 to C4 pathways. Are these conclusions justified?
First we consider analytic errors. Measurementerrorsin s13C lead directlyto errors
3X
1976
1978
1980
1982
1984
1986
d(O2/N2)/dt
d(CO2)/dt
CO2(in 1995)
1995
1996
1997 1998
Year
Fig. 2. The global average 813C record,measured by the NOAA-CMDL/CU-INSTAAR
networkwith calibrationas describedin the text.
1994
1996
1
x
A
0
1998
i
aredueto standarddrift
ind(O2/N2)/dt
Table1. Valuesof the quantitiesusedinthisstudy.Uncertainties
Thelatteris givenforan averageof datafromtwo sites.The"values"
andchoiceof fit (19),respectively.
of the measuredtrendsin 02 andCO2areillustrative
only,becauseourresultsarecalculatedfromrelated
quantities[see (19)].
Combustion
stoichiometry
stoichiometry
Photosynthetic
1994
1992
o
landandocean carbonstoragefromthis and otherstudies."Thisstudy (813C)"
Fig.3. Time-averaged
showsglobalaveragevaluescalculatedusingthe currentdata andthe methodof Ciaiset al. (9) with
G = 83.6 GtC%o/year.No errorsareshownfor 813CbecauseG is determinedusingO0/N2(see text).
andCapeGrimrecords.ThevaluesforSamoa
"Thisstudy(02)" showsthe averageof the Barrow/Alert
and Syowaare presentedwithouterrorestimatesbecausethe recordsare too short for meaningful
result(22) shown here has been adjustedto reflecta different
statisticalanalysis.The Langenfelds
averagingperiod(36).
[cement
1993
1990
1988
2
Year
[fuel
1992
0
0
-i
Value
Quantity
1991
m
4
d/dt
83m
13uel(in
813
1995)
(in 1995)
eair-land
Eair-sea
www.sciencemag.org SCIENCE VOL287
-6.21 + 0.37 GtC/year
-0.184
0.011 GtC/year
1.43+ 0.02
1.1 + 0.06
-16 + 0.8 + 0.35 per meg/year
1.24 _ 0.05ppm/year
760 ? 1 GtC
-0.013 + 0.008%o/year
-29.4 ? 1.8 %o
-7.86 0.015 %0
-18
+ 1%
-2 + 1 %o
31 MARCH2000
Source
(1)
(1)
(16)
(17)
This study
Thisstudy
Thisstudy
Thisstudy
(32)
Thisstudy
Thisstudy
(33, 34)
2469
REPORTS
i
i
i
Two-year
Fig. 4.
smoothed trends in
terrestrial and ocean- -...
.
.
ic carbon
uptake
based on 802/N2 and
s13C. The light lines
show the "global" average of fLand(upper
portion) and focean
(lower portion). The
meaning of "global"
changes as records
from new stations
02-based land flux. Average of 2, 3 & 4 stations
-;
-O
are added to the
513C-based land flux. Global average
average. Dark lines
show the global aver- oceanflux.Averageof 2, 3 & 4 stations
MgO2-based
age values for land
'
C13C-based ocean flux. Global average
- 4
and ocean carbon
on
the
based
storage
NOAA-CMDL/CU-IN3 o
STAAR s13C data set
2
(71) adjusted to the
_
CSIRO calibration, as
described in the text.
Carbon fluxes and
,
,
global average values
are calculated from
1997
1998 <
1995
1996
1991
1992
1993
1994
the 813C measureCalendarYear
ments according to
the methods of Ciais et al. (9). G has been adjusted to 83.6 GtC %o/year to optimize 813C-O2
agreement for the 6-year average fluxes (Fig. 3). 02-based records have been calculated as if
each station represented the entire atmosphere.
ablydo exist andcontributeat some level to the
fluctuationsshown in Fig. 4.
The agreement of the 13C and 602/N2
recordsis far from perfect.The 02-basedvariability inf,ce,,
and particularly infld
is greater
thanthatfoundin the gi3C-basedrecords.Some
dampingof the 8b3C-basedsignal is to be expected. Damping will occur because the 02
calculationsimplytreatsthe atmosphereas one
largebox, whereas813Ccalculationsuse a model of atmospherictransportto estimatethe effects of verticalmixing (9). Alternatively,the
smaller '3C-basedswings may reflectcovarying changes in carbon storage and terrestrial
isotopicdiscrimination.
Apparentdiscrepancies
at the ends of the recordsprobablyreflect the
fitting algorithm(28) and infrequentsampling
of
02/N2
in 1991.
As the smoothingperiodis decreasedbelow
the 2-yearvalue shown here, we see implausibly large variationsdevelop in the 02-based
carbon fluxes. These higher frequencyvariations are not due to fluctuationsin our standards,becausethese variationsat differentsites
arenot coherent.The most likely explanationis
that the 02 datapossess a significantregional
componentand that interannualvariationsin
ocean productionandventilationhave a significantimpacton atmospheric02 on time scales
of roughly 1 year.
With these reservationsin mind, it appears
that, as expected(29), the land biosphereis a
highly variablereservoir.Variabilitymust be
drivenby climate,butthe responseto forcingis
complex.AlthoughEl Niio has oftenbeenposited as a control of terrestrialcarbon storage
2470
(30), the multivariateEl Nifio index was high
duringour entireperiodof study,except for a
very weak La Nina episode from August 1995
to early 1997. Thus, other factorsmust be at
least partly responsiblefor the variabilitywe
observein our 6-yearrecord.
References and Notes
1. G. Marlandet al., in Fifth InternationalCarbonDioxide ConferenceExtendedAbstracts,R.Baum,I. Enting,
R. Francey,M. Hopkins,P. Holper, Eds. (CSIRODivision of Atmospheric Research, Aspendale, Victoria,
Australia,1997), p. 4.
2. One Gt = 1015 g; the addition of 1 Gt of carbon to
the atmosphere increases the amount of CO2 by
0.471 ppm.
3. L.Machta and E. Hughes, Science 168, 199 (1970).
4. R.F. Keelingand S. R.Shertz,Nature 358, 723 (1992).
5. C. D. Keeling,S. C. Piper,M. Heimann, in Aspects of
Climate Variabilityin the Pacific and WesternAmericas, D. H. Peterson, Ed.[Geophys. Monog. Am. Geophys. U. 55 (1989)], pp. 305-363.
6. FollowingKeelingand Shertz (4), we express changes
in atmospheric 02/Nz in per meg, defined as
8(O2/N2)=
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
/N
1Z/NZ
1
A change of 4.8 ppm is equivalent to 1 per meg,
because 02 makes up 20.9% of air.
R. F. Keeling et al., Global Biogeochem. Cyc. 7, 37
(1993).
M. Bender et al., Global Biogeochem. Cyc. 10, 9
(1996).
P. Ciais et al., J. Geophys. Res. 100, 5051 (1995).
M. Heimann and E. Maier-Reimer,Global Biogeochem. Cyc. 10, 89 (1996).
M. Trolieret al., J. Geophys. Res. 101, 25897 (1996).
K. Masarieet al., in preparation.
R. Franceyet al., Nature 373, 326 (1995).
P. Ciais et al., Science 269, 1098 (1995).
R. F. Keeling,S. C. Piper, M. Heimann, Nature 381,
218 (1996).
R. F. Keeling,thesis, HarvardUniversity (1988).
J. P. Severinghaus,thesis, ColumbiaUniversity(1995).
31 MARCH2000 VOL287
18. 802/N2 was measuredat the Universityof RhodeIsland
(URI)by mass spectrometry(8, 31), CO2was measured
(21). Smooth curves are fits to the
by NOAA/CMDL
data accordingto the algorithmof Thoninget al. (26).
The CO2 data points are analyses of the same flasks
analyzedfor O2/Nz.The COzfits shown here are made
to a separate set of analyses taken as part of the
NOAA/CMDL
cooperativeflask samplingnetwork(21).
The individualpointsshow the valuesused in determiningfLandand focean (see text). Each point is the average
of two flasks collected simultaneously.Flaskpairsare
discarded if they differ by more than 16 per meg.
Becausethe Keeling802/N2 data are measuredrelative
to a differentstandard,we adjustthose data by 66.7 per
meg [the mean differencebetween the URIand Scripps
Institutionof Oceanography(SIO)recordsat CapeGrim
duringthe 5-year periodof overlap].The adjustedSIO
data that predate URImeasurementsare appendedto
the beginningof the URIrecord,and fits are made to
the composite record.
19. To minimize errors in our calculation of focean we
combine measurementsof 02 and CO2for each flask
(Eq.4) and then calculatethe time-rate-of-changeof
this quantity.Inso doing,high-frequencyexcursionsin
02 due to localevents (inversions,pollution,etc.) will be
largely removed because of the inverse excursionsin
CO2.To determinetime-averagedocean carbonstorage
and CapeGrim,we calculate ocean and
at Barrow/Alert
fit a straightline to (i) the 1-yeartrend (26) and (ii)the
actual data values at the time of year when 0,2/N2
reaches a minimum at each station (an -10-week
interval starting on 16 Februaryand 21 August at
Barrowand CapeGrim,respectively).Resultsshown are
the averageof these two methods. We considercriterion (ii) in orderto select times when the deep mixed
layer makes it most likely that the ocean will be in
equilibriumwith the atmosphere and the effect of
ventilationimbalanceswill be minimized.
20. P. P. Tans, J. A. Berry,R. F. Keeling,Global Biogeochem. Cyc. 7, 353 (1993).
21. T.J. Conwayetal.,J. Geophys.Res. 99, 22831 (1994).
22. R. Langenfeldset al., Geophys. Res. Lett. 26, 1897
(1999).
23. M. Battle et al., Nature 383, 231 (1996).
24. P. Rayner,I. Enting,R. Francey,R. Langenfelds,Tellus
51, 213 (1999).
25. J. Sarmiento and E. Sundquist, Nature 356, 589
(1992).
26. K.W. Thoning,P. P. Tans,W. D. Komhyr,J. Geophys.
Res. 94, 8549 (1989).
27. R. Franceyet al., NOAATech. Rep. 22 (1994).
28. For fitting purposes, the first year of each record is
taken as representative of the previous 2 years of
(nonexistant) data. Similarlyfor the last year of data.
Thus, short-term fluctuations at the beginnings and
ends of the records can greatly distort the fit.
29. D. S. Schimel, Global Change Biol. 1, 77 (1995).
30. C. Keelinget al., Nature 375, 666 (1995).
31. M. L. Bender et al., Geochim. Cosmochim.Acta 58,
4751 (1995).
32. R.J. Andres,personal communication.
33. R. Wanninkhof,Tellus 37B, 128 (1985).
34. U. Sieghethaler and K. MOnnich,in Carbon Cycle
Modeling, B. Bolin, Ed. (Wiley, New York,1981), pp.
249-257.
35. Samples were collected from the following stations:
Alert (82?28'N, 62?30'W), Barrow (79?19'N,
156?36'W), AmericanSamoa (14?15'S, 170034'W),
Cape Grim (40?41'S, 144?41'E), Baring Head
(41"20'S, 174?50'E), and Syowa (69?00'S, 39"35'E).
36. The result of Langenfeldset al. applied to the years
1978-1997. Using the results from this work for the
period 1991.5-1997, we infer the values for the
period 1978-1991.5. This is the quantity shown in
Fig.3. Errorsfor the Langenfeldset al. result and our
work are highly correlated because of common data
and methods.
37. We thank the staffs of the various observatoriesfor
their carefulcollectionof airsamplesand R.Keeling,R.
Langenfelds,and M. Heimannfor helpful discussions.
Supportedby grantsfrom NSF,the U.S.Environmental
ProtectionAgency,and the NOAAGlobalChange Research Program.
13 September 1999; accepted 22 February2000
SCIENCEwww.sciencemag.org