Content uploaded by Gary D. Clow
Author content
All content in this area was uploaded by Gary D. Clow on Nov 12, 2019
Content may be subject to copyright.
DOI: 10.1126/science.282.5387.268
, 268 (1998);282 Science , et al.D. Dahl-Jensen
Past Temperatures Directly from the Greenland Ice Sheet
This copy is for your personal, non-commercial use only.
clicking here.colleagues, clients, or customers by , you can order high-quality copies for yourIf you wish to distribute this article to others
here.following the guidelines can be obtained byPermission to republish or repurpose articles or portions of articles
): February 3, 2011 www.sciencemag.org (this infomation is current as of
The following resources related to this article are available online at
http://www.sciencemag.org/content/282/5387/268.full.html
version of this article at: including high-resolution figures, can be found in the onlineUpdated information and services,
http://www.sciencemag.org/content/282/5387/268.full.html#ref-list-1
, 2 of which can be accessed free:cites 20 articlesThis article
339 article(s) on the ISI Web of Sciencecited by This article has been
http://www.sciencemag.org/content/282/5387/268.full.html#related-urls
40 articles hosted by HighWire Press; see:cited by This article has been
http://www.sciencemag.org/cgi/collection/geochem_phys
Geochemistry, Geophysics subject collections:This article appears in the following
registered trademark of AAAS. is aScience1998 by the American Association for the Advancement of Science; all rights reserved. The title CopyrightAmerican Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005.
(print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by theScience
on February 3, 2011www.sciencemag.orgDownloaded from
nated to a surface Al, providing a mechanism
for the interchange of O
s
and O
ads
. These re-
sults also provide evidence for incipient
Al(OH)
3
formation on the surface. The ultimate
structure of the heavily hydrated surface is
clearly very complicated and may depend
strongly on sample history. It is possible that
the Al(OH)
3
species can be removed complete-
ly (perhaps starting near steps or other defects),
leaving a less reactive surface that is completely
O
s
H-terminated, which is similar to the known
surfaces of aluminum hydroxides (1).
An idealized model for fully hydroxylated
a-Al
2
O
3
(0001) (28) replaces each surface Al
with three H atoms (Fig. 5A), yielding a
coverage .15 OH per square nanometer.
Room-temperature MD simulations of this
model revealed a complex dynamic structure
(Fig. 5B), with one out of every three OH
groups, on average, lying parallel to the sur-
face because of in-plane hydrogen bonding.
Calculated O–H vibrational spectra (25)
yielded two broad peaks at ;3470 and 3650
cm
–1
, with the peak at ;3470 cm
–1
corre-
sponding to in-plane OH groups. The peak at
;3650 cm
–1
is close to the single peak (3720
to 3733 cm
–1
) that is observed in most mea-
surements on hydroxylated a-Al
2
O
3
(0001)
(29) and to the range that is generally as-
signed to bridging OH groups (2,26). The
peak at ;3470 cm
–1
is red-shifted by hydro-
gen bonding and is generally not seen in
single-crystal experiments, perhaps because
of selection rules or because it is too broad.
Our finding of two peaks split by 200 cm
–1
contradicts all previous classifications of OH
vibrations (and subsequent cluster modeling)
(2) on aluminas, which assume that all OH
groups with the same coordination of O and
neighboring Al have the same frequency. By
this criterion, all of the surface OH groups in
Fig. 5 are equivalent; however, their stretch
frequencies clearly depend also on longer
range environmental effects.
The present investigation of a-Al
2
-
O
3
(0001) has elucidated several aspects of
the complex interactions of H
2
O with an
alumina surface, especially the dynamics of
dissociation reactions at low and high cover-
ages. On the basis of these results, a consis-
tent interpretation of a diverse set of experi-
mental data on hydroxylated alumina surfac-
es begins to emerge.
References and Notes
1. K. Wefers and C. Misra, Alcoa Tech. Pap. 19 (revised)
(Alcoa Laboratories, St. Louis, MO, 1987).
2. H. Kno¨zinger and P. Ratnasamy, Catal. Rev. Sci. Eng.
17, 31 (1978).
3. M. Gautier et al.,J. Am. Ceram. Soc. 77, 323 (1994).
4. P. de Sainte Claire, K. C. Hass, W. F. Schneider, W. L.
Hase, J. Chem. Phys. 106, 7331 (1997).
5. G. N. Robinson, Q. Dai, A. Freedman, J. Phys. Chem. B
101, 4940 (1997).
6. R. Car and M. Parrinello, Phys. Rev. Lett. 55, 2471
(1985).
7. The gradient-corrected exchange-correlation [Bern-
stein, Lee, Yang, and Primakoff (BLYP)] functional
used here is from A. D. Becke [Phys. Rev. A 38, 3098
(1988)] and C. Lee, W. Yang, and R. Parr [Phys. Rev. B
37, 785 (1988)]. Norm-conserving numerical pseudo-
potentials were generated for Al and O with the
procedure of N. Troullier and J. L. Martins [ibid. 43,
1993 (1991)], and a local analytic pseudopotential
was derived for H. This is essentially a softened
Coulomb potential with a core radius of 0.25 atomic
units. Electron wave functions are expanded in a
plane-wave basis set with an energy cutoff of 70
rydbergs (Ry). We used the Car-Parrinello Molecular
Dynamics code in the parallelized 2.5 version (devel-
oped by J. Hutter and copyrighted by IBM, Armonk,
NY). All calculations were performed on a 32-node
IBM RS6000 SP at the IBM Watson Research Labora-
tory (Yorktown Heights, NY ).
8. In the MD runs, a value of 400 au was used for the
fictitious electron mass of the Car-Parrinello Lagrang-
ian multipliers (6), and each hydrogen molecule was
replaced by deuterium to improve the separation
between electronic and ionic degrees of freedom. The
time step in the Verlet algorithm for the integration
of the equations of motions was ;0.1 fs.
9. The importance of chemical reaction dynamics in
general has recently been highlighted in a special
issue of Science [Reaction Dynamics, Science 279,
1875–1895 (1998)].
10. A. Curioni et al.,J. Am. Chem. Soc. 119, 7218 (1997).
11. V. E. Puchin et al.,Surf. Sci. 370, 190 (1997); J. Ahn
and J. W. Rabalais, ibid. 388, 121 (1997).
12. See, for example, S. Blonski and S. H. Garofalini, ibid.
295, 263 (1993).
13. See, for example, M. Causa, R. Dovesi, C. Pisani, C.
Roetti, ibid. 215, 259 (1989); I. Manassidis, A. De
Vita, M. J. Gillan, Surf. Sci. Lett. 285, L517 (1993); I.
Frank, D. Marx, M. Parrinello, J. Chem. Phys. 104,
8143 (1996).
14. J. M. McHale, A. Auroux, A. J. Perrotta, A. Navrotsky,
Science 277, 788 (1997). For earlier work, see P. A.
Thiel and T. E. Madey, Surf. Sci. Rep. 7, 211 (1987)
and references therein.
15. J. M. Wittbrodt, W. L. Hase, H. B. Schlegel, J. Phys.
Chem. B 102, 6539 (1998).
16. K. C. Hass, W. F. Schneider, A. Curioni, W. Andreoni, in
preparation.
17. Earlier calculations used much smaller supercells than
the present work. Such studies were therefore limited
in their ability to provide accurate adsorbate struc-
tures and energies and to study the H
2
O coverage
dependence and phenomena such as collective ef-
fects and surface diffusion.
18. J. Goniakowski and M. J. Gillan, Surf. Sci. 350, 145
(1996); P. J. D. Lindan, N. M. Harrison, J. M. Hold-
ender, M. J. Gillan, Chem. Phys. Lett. 261, 246 (1996);
P. J. D. Lindan, N. M. Harrison, M. J. Gillan, Phys. Rev.
Lett. 80, 762 (1998).
19. W. Langel and M. Parrinello, J. Chem. Phys. 103, 3240
(1995).
20. Lagrange multipliers were introduced to constrain
the relevant H–O
s
distance, and the average con-
straint forces were determined from constant tem-
perature simulations [S. Nose´, J. Chem. Phys. 81, 511
(1984); W. G. Hoover, Phys. Rev. A 31, 1695 (1985)]
of at least 0.2 ps.
21. S. Scheiner, in Proton Transfer in Hydrogen-Bonded
Systems, T. Bountis, Ed. (Plenum, New York, 1992), p.
29.
22. S. Blonski and S. H. Garofalini, J. Phys. Chem. 100,
2201 (1996).
23. The temperature was not controlled but was in-
creased slowly from ;100 to ;350 K. The system
was then allowed to evolve for a time interval of .1
ps. The average temperature was 250 K.
24. D. E. Brown, D. J. Moffatt, R. A. Wolkow, Science 279,
542 (1998).
25. Vibrational frequencies were estimated from the
power spectra of the (partial) velocity-velocity auto-
correlation functions and were rescaled to account
for the fictitious electronic mass and the different
mass used for the proton.
26. V. I. Lygin and I. S. Muzyka, Russ. J. Phys. Chem. 69,
1829 (1995); A. Tsyganenko and P. Mardilovich, J.
Chem. Soc. Faraday Trans. 92, 4843 (1996).
27. B. A. Huggins and P. D. Ellis, J. Am. Chem. Soc. 114,
2098 (1992).
28. M. A. Nygren, D. H. Gay, C. R. A. Catlow, Surf. Sci.
380, 113 (1997).
29. C. Morterra, G. Ghiotti, E. Garrone, F. Boccuzzi, J.
Chem. Soc. Faraday Trans. 1 72, 2722 (1976); J. G.
Chen, J. E. Crowell, J. T. Yates, J. Chem. Phys. 84, 5906
(1986); V. Coustet and J. Jupille, Surf. Sci. 307, 1161
(1994).
11 August 1998; accepted 3 September 1998
Past Temperatures Directly
from the Greenland Ice Sheet
D. Dahl-Jensen,* K. Mosegaard, N. Gundestrup, G. D. Clow,
S. J. Johnsen, A. W. Hansen, N. Balling
A Monte Carlo inverse method has been used on the temperature profiles
measured down through the Greenland Ice Core Project (GRIP) borehole, at the
summit of the Greenland Ice Sheet, and the Dye 3 borehole 865 kilometers
farther south. The result is a 50,000-year-long temperature history at GRIP and
a 7000-year history at Dye 3. The Last Glacial Maximum, the Climatic Optimum,
the Medieval Warmth, the Little Ice Age, and a warm period at 1930 A.D. are
resolved from the GRIP reconstruction with the amplitudes –23 kelvin, 12.5
kelvin, 11 kelvin, –1 kelvin, and 10.5 kelvin, respectively. The Dye 3 temper-
ature is similar to the GRIP history but has an amplitude 1.5 times larger,
indicating higher climatic variability there. The calculated terrestrial heat flow
density from the GRIP inversion is 51.3 milliwatts per square meter.
Measured temperatures down through an ice
sheet relate directly to past surface tempera-
ture changes. Here, we use the measurements
from two deep boreholes on the Greenland
Ice Sheet to reconstruct past temperatures.
The GRIP ice core (72.6°N, 37.6°W) was
successfully recovered in 1992 (1, 2), and the
3028.6-m-deep liquid-filled borehole with a
diameter of 13 cm was left undisturbed. Tem-
peratures were then measured down through
the borehole in 1993, 1994, and 1995 (3, 4).
We used the measurements from 1995 (Fig.
1) (4), because there was no remaining evi-
dence of disturbances from the drilling and
REPORTS
9 OCTOBER 1998 VOL 282 SCIENCE www.sciencemag.org268
on February 3, 2011www.sciencemag.orgDownloaded from
the measurements were the most precise (65
mK). Temperatures measured in a thermally
equilibrated shallow borehole near the drill
site are used for the top 40 m, because they
are more reliable than the GRIP profile over
this depth (5). The present mean annual sur-
face temperature at the site is –31.70°C. The
2037-m-deep ice core from Dye 3 (65.2°N,
43.8°W) was recovered in 1981. We used
temperature data from 1983 measurements
with a precession of 30 mK (6, 7). The tem-
peratures at the bedrock are –8.58°C at GRIP
and –13.22°C at Dye 3. Calculations show
that the basal temperatures have been well
below the melting point throughout the past
100,000 years (8). Because there are still
climate-induced temperature changes near
the bedrock, we included 3 km of bedrock in
the heat flow calculatin.
Past surface temperature changes are in-
dicated from the shape of the temperature
profiles (Fig. 1). We used a coupled heat- and
ice-flow model to extract the climatic infor-
mation from the measured temperature pro-
files. The temperatures down through the ice
depend on the geothermal heat flow density
(heat flux), the ice-flow pattern, and the past
surface temperatures and accumulation rates.
The past surface temperatures and the geo-
thermal heat flow density are unknowns,
whereas the past accumulation rates and ice-
flow pattern are assumed to be coupled to the
temperature history through relations found
from ice-core studies (9–11). The total ice
thickness is assumed to vary 200 m as de-
scribed in (9). The coupled heat- and ice-flow
equation is (7, 9, 12)
rc]T
]t5¹z~K¹T!2rcnYz¹T1f
where T(x,z,t) is temperature, tis time, zis
depth, xis horizontal distance along the flow
line, r(z) is ice density, K(T,r) the thermal
conductivity, c(T) is the specific heat capac-
ity, and f(z) is the heat production term. The
ice velocities, vY(x,z,t), are calculated by an
ice-flow model (9, 13). Model calculations to
reproduce a present-day temperature profile
through the ice sheet are started 450,000
years ago (ka) at GRIP (100 ka at Dye 3),
D. Dahl-Jensen, K. Mosegaard, N. Gundestrup, S. J.
Johnsen, A. W. Hansen, Niels Bohr Institute for As-
tronomy, Physics and Geophysics, Department of
Geophysics, Juliane Maries Vej 30, DK-2100 Copen-
hagen OE, Denmark. G. D. Clow, USGS-Climate Pro-
gram, Box 25046, MS 980, Denver Federal Center,
Denver, CO 80225, USA. N. Balling, Department of
Earth Sciences, Geophysical Laboratory, University of
Aarhus, Finlandsgade 8, DK-8200 Aarhus N, Denmark.
*To whom correspondence should be addressed. E-
mail: ddj@gfy.ku.dk
020 0
1000
500
1000
1000
2000
3000 2000
1500
1000
500
0
-30
-31.8 -20 -18
-31.6 -31.4
CO CO
LIA LIA
-20 -20 -18 -16 -14-10
Temperature (°C)
A
B
C
D
Temperature (°C)
Depth (m)
Depth (m)
GRIP Temperature Profile 1995 Dye 3 Temperature Profile 1983 Fig. 1. The GRIP and Dye 3 temperature profiles
[blue trace in (A) and (C)] are compared to
temperature profiles [red trace in (A) and (C)]
calculated under the condition that the present
surface temperatures and accumulation rates
have been unchanged back in time. (A) The
GRIP temperature profile measured in 1995.
The cold temperatures from the Glacial Period
(115 to 11 ka) are seen as cold temperatures
between 1200- to 2000-m depth. (B) The top
1000 m of the GRIP temperature profiles are
enlarged so the Climatic Optimum (CO, 8 to 5
ka), the Little Ice Age (LIA, 1550 to 1850 A.D.),
and the warmth around 1930 A.D. are indicated
at the depths around 600, 140, and 60 m,
respectively. (C) The Dye 3 temperature profile
measured in 1983. Note the different shape of
the temperature profiles when compared to
GRIP and the different depth locations of the
climate events. (D) The top 1500 m of the Dye
3 temperature profiles are enlarged so the CO,
the LIA, and the warmth around 1930 A.D. are
indicated at the depths around 800, 200, and
70 m, respectively.
700
600
500
400
300
200
100
00
100
250
200
150
100
50
00
100
200
300
400
500
600
Temperature (°C)
A
DE
BCF
Heat flux (mW/m2)
Geothermal heat flux1000 A.D.1600 A.D.1970 A.D.
Number of solutionsNumber of solutions
8 ka BP 25 ka BP
200
300
400
500
0
100
200
300
400
500 450
400
350
300
250
200
150
100
50
0
-32.5 -32 -31.5 -31 -30.5 -33 -32 -31 -31.5 -31 -30.5 -30 -29.5
-29.5-30-30.5-31-31.5 50.5 51 51.5 52 52.5
-51-52-53-54-55-56-57-58-59
Fig. 2. (Athrough E) The probability distributions of the past surface temperatures at the
Greenland Ice Sheet summit at selected times before present. They are constructed as histograms
of the 2000 Monte Carlo sampled and accepted temperature histories (17). All temperature
distributions are seen to have a zone with maximum values, the most likely values, which are
assumed to be the reconstructed surface temperature at these times (18). (F) The probability
distribution of the sampled geothermal heat flow densities. The most likely value is 51.3 mW/m
2
.
REPORTS
www.sciencemag.org SCIENCE VOL 282 9 OCTOBER 1998 269
on February 3, 2011www.sciencemag.orgDownloaded from
more than twice the time scale for thermal
equilibrium of the ice-bedrock, so the un-
known initial conditions are forgotten when
generating the most recent 50,000-year tem-
perature history (7000 years for Dye 3).
We developed a Monte Carlo method to fit
the data and infer past climate. The Monte
Carlo method tests randomly selected combina-
tions of surface temperature histories and geo-
thermal heat flow densities by using them as
input to the coupled heat- and ice-flow model
and considering the resulting degrees of fit be-
tween the reproduced and measured tempera-
ture profiles (14–16). Our results for each site
are based on tests of 3.3 310
6
combinations
of temperature histories and heat flow densities,
of which 2000 solutions have been selected
(17). The 2000 temperature histories and heat
flow densities are sampled with a frequency
proportional to their likelihood (14, 15), and all
accepted solutions fit the observations within
their limits of uncertainty.
Histograms of the sampled geothermal heat
flow densities and of the temperature histories
at each time before present can be made (for
example, Fig. 2). The distributions in general
show that there is a most likely value, a maxi-
mum, at all times, which we refer to as the
temperature history (18). The distribution of
accepted geothermal heat flow densities (Fig.
2F) has a median of 51.3 60.2 mW/m
2
, which
is slightly higher than the heat flow density
from Archean continental crust across the Baf-
fin Bay in Canada. A few heat flow measure-
ments have been made from the coast of Green-
land (36 and 43 mW/m
2
), but these are not
corrected for long-term climate variations and
are minimum values (19). The homogeneous
thermal structure of ice is an advantage when
the heat flow density and the temperature his-
tory are to be reconstructed (20).
Histograms from the GRIP reconstruction
(Fig. 3) show that temperatures at the Last
Glacial Maximum (LGM) were 23 62K
colder than at present (21). The temperatures
at this time, 25 ka, reflect the cold temperatures
seen on the measured temperature profile at a
depth of 1200 to 2000 m. Alternative recon-
structions of the ice thickness and accumulation
rates all reproduce LGM temperatures within 2
K(9, 10, 22, 23). The cold Younger Dryas and
the warm Bølling/Allerød periods (24) are not
resolved in the inverse reconstruction. The tem-
perature signals of these periods have been
obliterated by thermal diffusion because of their
short duration (25). After the termination of the
glacial period, temperatures in our record in-
crease steadily, reaching a period 2.5 K warmer
than present during what is referred to as the
Climatic Optimum (CO), at 8 to 5 ka. Follow-
ing the CO, temperatures cool to a minimum of
0.5 K colder than the present at around 2 ka.
The record implies that the medieval period
around 1000 A.D. was 1 K warmer than present
in Greenland. Two cold periods, at 1550 and
1850 A.D., are observed during the Little Ice
Age (LIA) with temperatures 0.5 and 0.7 K
below the present. After the LIA, temperatures
reach a maximum around 1930 A.D.; tempera-
tures have decreased during the last decades
(26). The climate history for the most recent
times is in agreement with direct measurements
in the Arctic regions (27). The climate history
for the last 500 years agrees with the general
understanding of the climate in the Arctic re-
gion (28) and can be used to verify the temper-
ature amplitudes. The results show that the
temperatures in general have decreased since
the CO and that no warming in Greenland is
observed in the most recent decades.
As seen in Fig. 3, resolution decreases back
AB
C
Past temperatures (°C)
Years before present (ka) Years before present (ka)
Years A.D.
Fig. 3. The contour plots of all the GRIP temperature histograms as a function of time describes the
reconstructed temperature history (red curve) and its uncertainty. The temperature history is the
history at the present elevation (3240 m) of the summit of the Greenland Ice Sheet (21). The white
curves are the standard deviations of the reconstruction (18). The present temperature is shown as
a horizontal blue curve. The vertical colored bars mark the selected times for which the temper-
ature histograms are shown in Fig. 2. (A) The last 100 ky BP. The LGM (25 ka) is seen to have been
23 K colder than the present temperature, and the temperatures are seen to rise directly into the
warm CO 8 to 5 ka. (B) The last 10 ky BP. The CO is 2.5 K warmer than the present temperature,
and at 5 ka the temperature slowly cools toward the cold temperatures found around 2 ka. (C) The
last 2000 years. The medieval warming (1000 A.D.) is 1 K warmer than the present temperature,
and the LIA is seen to have two minimums at 1500 and 1850 A.D. The LIA is followed by a
temperature rise culminating around 1930 A.D. Temperature cools between 1940 and 1995.
-30
-32
-34
86420
200015001000500
-31
-17.5
-20.0
-22.5
Dye 3
GRIP
-18
-19
-20
-21
Years before present (ka)
A
B
Years A.D.
Dye 3 past surface temperatures (°C)
GRIP past surface temperatures (°C)
-32
Fig. 4. The reconstructed temperature histories
for GRIP (red curves) and Dye 3 (blue curves) are
shown for the last 8 ky BP (A) and the last 2 ky BP
(B). The two histories are nearly identical, with
50% larger amplitudes at Dye 3 than found at
GRIP. The reconstructed climate must represent
events that occur over Greenland, probably the
high-latitude North Atlantic region.
REPORTS
9 OCTOBER 1998 VOL 282 SCIENCE www.sciencemag.org270
on February 3, 2011www.sciencemag.orgDownloaded from
in time (25, 29). For the GRIP reconstruction,
an event with a duration of 50 years and an
amplitude of 1 K can be resolved 150 years
back in time with a measurement accuracy of 5
mK; an event with a similar amplitude but a
duration of 1000 years can be detected back to
5 ka. An event that occurred 50 ka will now be
observed in the temperature profile at the bed-
rock. Climate events for times older than
50,000 years before present (ky BP) are not
well resolved (30). At Dye 3, the reconstructed
climate history extends only to 7 ka, because
the ice is 1000 m thinner than at the summit and
surface accumulation rate is 50% higher. The
LGM is not well resolved in the Dye 3 record,
and consequently the geothermal heat flow den-
sity is not uniquely determined (31). On the
other hand, the recent climate history has a
higher resolution because of the increased ac-
cumulation (Fig. 4).
The Dye 3 record is nearly identical with the
GRIP record back to 7 ka, but the amplitudes
are 50% higher. Thus, the resolved climate
changes have taken place on a regional scale;
many are seen throughout the Northern Hemi-
sphere (27, 28, 32). GRIP is located 865 km
north of Dye 3 and is 730 m higher in elevation.
Surface temperatures at the summit are influ-
enced by maritime air coming in from the North
Atlantic and air masses arriving from over
northeastern Canada (associated with the Baffin
trough) (28, 32, 33). Temperatures at Dye 3 will
be influenced to a greater degree by the North
Atlantic maritime air masses. Dye 3 is located
closer to the center of the highest atmospheric
variability, which is associated with large in-
terseasonal, interannual, and decadal tempera-
ture changes (32, 34). It is therefore believed
that the observed difference in amplitudes be-
tween the two sites is a result of their different
geographic location in relation to variability of
atmospheric circulation, even on the time scale
of a millennium.
References and Notes
1. Greenland Ice-Core Project (GRIP) members, Nature
364, 203 (1993).
2. W. Dansgaard et al.,ibid., p. 218.
3. N. S. Gundestup, D. Dahl-Jensen, S. J. Johnsen, A.
Rossi, Cold Reg. Sci. Technol. 21, 399 (1993).
4. G. D. Clow, R. W. Saltus, E. D. Waddington, J. Glaciol.
42, 576 (1996).
5. The deep borehole is located in a building, and the liquid
surface in the borehole is found at a depth of 40 m. The
temperatures measured in the top 40 m are very dis-
turbed, so we used measurements from an air-filled
shallow borehole (100 m) near the borehole.
6. N. S. Gundestrup and B. L. Hansen, J. Glaciol. 30, 282
(1984).
7. D. Dahl-Jensen and S. J. Johnsen, Nature 320, 250
(1986).
8. D. Dahl-Jensen et al.,J. Glaciol. 43, 300 (1997).
9. Between 50 and 20 ka, the ice thickness was 50 m
less than at present, even though the ice sheet cov-
ered a larger area. The maximum ice thickness of
3230 m is found at 10 ka, after which the ice thick-
ness gradually has decreased to the present 3028.6
m. The depression and uplift of the bedrock influenc-
es the elevation of the surface [S. J. Johnsen, D.
Dahl-Jensen, W. Dansgaard, N. S. Gundestrup, Tellus
B47, 624 (1995)].
10. K. M. Cuffey and G. D. Clow, J. Geophys. Res. 102,
26383 (1997).
11. The past accumulation rates are determined by cou-
pling them to the past (unknown) temperature
through the relation l(T)5l
O
exp[0.0467(T–T
O
)–
0.000227(T–T
O
)
2
], where l(T) is the accumulation
rate at the surface temperature T,l
O
is the present
ice accumulation rate, which is 0.23 m/year at GRIP
and 0.49 m/year at Dye 3, and T
O
is the present
surface temperatures at the sites: –31.7°C at GRIP
and –20.1°C at Dye 3, respectively (9).
12. S. J. Johnsen, IAHS-AISH Publ. 118, 388 (1977).
13. S. J. Johnsen and W. Dansgaard, NATO ASI Ser. I
Global Environ. Change 2, 13 (1992).
14. K. Mosegaard and A. Tarantola, J. Geophys. Res. 100,
12431 (1995).
15. K. Mosegaard, Inverse Problems 14, 405 (1998).
16. Our Monte Carlo scheme is a random walk in the high
dimensional space of all possible models, m (tempera-
ture histories and geothermal heat flow densities). The
temperature history has been divided in 125 intervals
(interval length is 25 ky at 450 ka and 10 years at
present). Including the geothermal heat flow density as
an unknown the model space is 126-dimensional. In
each step of the random walk, a perturbed model, m
pert
i
of the current model vector m
i
is proposed. The next
model becomes equal to m
pert
i
with an acceptance
probability P
accept
5min{1,exp(–[S(m
pert
i
)–S(m
i
)])},
where S(m)5S
j
(g
j
(m)–d
obs
j
)
2
, which is the misfit
function measuring the difference between g(m), the
calculated borehole temperatures, and d
obs
, the ob-
served temperatures. If the perturbated model is reject-
ed, the next model becomes equal to m
i
and a new
perturbed model is proposed. To ensure an efficient
sampling of all possible models, we developed ways of
choosing the temperature histories and geothermal
heat flow densities to be tested. The main scheme to
perturb the models is to randomly select one of the 126
temperature/heat flow density parameters and change
its value to a new value chosen uniformly at random
within a given interval. A singular value decomposition
(SVD) of the matrix G5{]g
j
/]m
i
}, evaluated in a
near-optimal model, yields a set of eigenvectors in the
model space whose orientations reveal efficient direc-
tions of perturbation for the random walk. The SVD
method is included as a possible method of perturbing
models especially in the start of the process as it speeds
the Monte Carlo scheme significantly.
17. Of the 3.3 310
6
models tested during the random
walk 30% have been accepted by the Monte Carlo
scheme (16). Every 500 is chosen of those where the
misfit function S (16) is less than the variance of the
observations. The waiting time of 500 has been cho-
sen to exceed the maximum correlation length of the
output model parameters. This is a necessary condi-
tion for the 2000 models to be uncorrelated. To
further ensure that the output models were uncor-
related, the random walk was frequently restarted at
several random selected points in the model space.
18. The probabilistic formulation of the inverse problem
leads to definition of a probability distribution in the
model space, describing the likelihood of possible
temperature histories and geothermal heat flow den-
sities. The Monte Carlo scheme is constructed to
sample according to this probability distribution. The
histograms in Fig. 2 describe the probability distribu-
tion of the geothermal heat flow density and tem-
peratures at times before present. The maxima in the
histograms thus describe the most likely values. The
method does not constrain the distributions to have
a single maximum, indeed there could be histograms
with several maxima, reflecting that more than one
value of the temperature at this time would give a
good fit to the observed temperature in the borehole.
The histograms however, are all seen to have a
well-defined zone with most likely past tempera-
tures. A soft curve is fitted to the histograms and the
maximum value is taken as the most likely value. The
standard deviations shown in Fig. 3 are derived as
deviations from the maximum value.
19. J. H. Sass, B. L. Nielsen, H. A. Wollenberg, R. J. Munroe,
J. Geophys. Res. 77, 6435 (1972).
20. C. Clauser et al.,ibid. 102, 18417 (1997); L. Guillou-
Frottier, J.-C. Marescal, J. Musset, ibid. 103, 7385
(1998); H. N. Pollack, S. J. Hurter, J. R. Johnson, Rev.
Geophys. 31, 267 (1993); W. G. Powell, D. S. Chap-
man, N. Balling, A. E. Beck, in Handbook of Terrestrial
Heat-Flow Density Determination (Kluwer Academic,
New York, 1988), pp. 167–222.
21. In order to produce a past temperature record from
the calculated past surface temperatures, the tem-
peratures have been corrected to the present eleva-
tion of the GRIP site (and Dye 3 site respectively)
using the surface elevation changes described in (9)
and a lapse rate of 0.006 K/m.
22. K. M. Cuffey et al.,Science 270, 455 (1995).
23. D. Dahl-Jensen, in Proceedings of the Interdisciplinary
Inversion Workshop 2, Copenhagen, 19 May 1993, K.
Mosegaard, Ed. (The Niels Bohr Institute for Astron-
omy, Physics and Geophysics, University of Copen-
hagen, Copenhagen, 1993), pp. 11–14.
24. C. U. Hammer et al.,Report on the stratigraphic
dating of the GRIP Ice Core.Special Report of the
Geophysical Department (Niels Bohrs Institute for
Astronomy, Physics and Geophysics, University of
Copenhagen, Copenhagen, in press).
25. J. Firestone, J. Glaciol. 41, 39 (1995).
26. The amplitude of the warming at 1930 A.D. must be
considered to be more uncertain. The information
leading to this result are the measured temperatures
in an open shallow borehole, where air movements
could influence the measurements.
27. D. Fisher et al.,NATO ASI Ser. I Global Environ.
Change 41, 297 (1996); J. W. C. White et al.,J.
Geophys. Res. 102, 26425 (1997); J. W. Hurrel, Sci-
ence 269, 676 (1995); P. Frich et al.,inDMI Scientific
Report 96-1 (Danish Meteorological Institute, Copen-
hagen, 1996).
28. R. G. Barry and R. J. Charley, Atmosphere, Weather &
Climate (Routledge, London, ed. 6, 1992); J. Overpeck
et al.,Science 278, 1251 (1997); H. H. Lamb, Climate
History and the Modern World (Routledge, London,
ed. 2, 1995); N. W. T. Brink and A. Weidick, Quat. Res.
4, 429 (1974).
29. G. D. Clow, Palaeogeogr. Palaeoclimatol. Palaeoecol.
98, 81 (1992).
30. To comply with this resolution the time steps have
been chosen with increasing length back in time. The
increasing length of the time steps can be considered
as an efficient way of calculating the mean temper-
atures in the intervals so full available resolution is
kept but the calculations are rationalized.
31. In (7), it is argued that parameter combinations of
mean glacial temperature, mean glacial accumula-
tion, and geothermal heat flow density can be found
that fit the Dye 3 measurements due to the reduced
resolution of the climate history reaching further
back than 7 ka. A combination with a geothermal
heat flow density of 38.7 mW/m
2
was chosen corre-
sponding to a mean glacial temperature 12 K colder
than the present temperatures. If a value of 51
mW/m
2
is chosen as that found for our inversion, the
mean glacial temperature is 19 K colder than the
present, which is well in agreement with the results
found for the GRIP reconstruction. Comparison of the
Dye 3 temperature history presented in (7) and that
presented here shows a general good agreement for
the last 7 ky. The history presented in (7) is more
intuitive and less detailed, and the history has not
been corrected for elevation changes. The ice thick-
ness was assumed constant in this reconstruction.
32. L. K. Barlov, J. C. Rogers, M. C. Serreze, R. C. Barry, J.
Geophys. Res. 102, 26333 (1997).
33. R. A. Keen, Occas. Pap. 34 (Institute of Arctic and
Alpine Research, University of Colorado, Boulder,
1980).
34. S. Shubert, W. Higgins, C. K. Park, S. Moorthi, M.
Svarez, An Atlas of ECMWF Analyses (1980–87). Part
II: Second Moment Quantities,NASA Tech. Memoran-
dum 100762 (1990); F. Rex, World Surv. Climatol. 4,
1 (1969).
35. This is a contribution to the Greenland Ice Core
Project (GRIP), a European Science Foundation pro-
gram with eight nations and the European Economic
Commission collaborating to drill through the central
part of the Greenland Ice Sheet. G.D.C. thanks the
USGS Climate History Program and NSF for support.
16 June 1998; accepted 1 September 1998
REPORTS
www.sciencemag.org SCIENCE VOL 282 9 OCTOBER 1998 271
on February 3, 2011www.sciencemag.orgDownloaded from
