A Large and Persistent Carbon Sink in the World's Forests

Abstract

The terrestrial carbon sink has been large in recent decades, but its size and location remain uncertain. Using forest inventory
data and long-term ecosystem carbon studies, we estimate a total forest sink of 2.4 ± 0.4 petagrams of carbon per year (Pg
C year–1) globally for 1990 to 2007. We also estimate a source of 1.3 ± 0.7 Pg C year–1 from tropical land-use change, consisting of a gross tropical deforestation emission of 2.9 ± 0.5 Pg C year–1 partially compensated by a carbon sink in tropical forest regrowth of 1.6 ± 0.5 Pg C year–1. Together, the fluxes comprise a net global forest sink of 1.1 ± 0.8 Pg C year–1, with tropical estimates having the largest uncertainties. Our total forest sink estimate is equivalent in magnitude to the
terrestrial sink deduced from fossil fuel emissions and land-use change sources minus ocean and atmospheric sinks.

Full text

DOI: 10.1126/science.1201609
, 988 (2011);333 Science , et al.Yude Pan
A Large and Persistent Carbon Sink in the World's Forests
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
): August 18, 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/333/6045/988.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/suppl/2011/07/13/science.1201609.DC1.html
can be found at: Supporting Online Material
http://www.sciencemag.org/content/333/6045/988.full.html#ref-list-1
, 11 of which can be accessed free:cites 33 articlesThis article
http://www.sciencemag.org/cgi/collection/geochem_phys
Geochemistry, Geophysics http://www.sciencemag.org/cgi/collection/atmos
Atmospheric Science subject collections:This article appears in the following
registered trademark of AAAS. is aScience2011 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 August 18, 2011www.sciencemag.orgDownloaded from

propagation are directly accessible to anyone
with basic statistical knowledge. This should ul-
timately open the way for a complete character-
ization of the roles of direct and indirect top-down
and bottom-up mechanisms involved in the reg-
ulation of parasite densities (fig. S12 and table S1)
in the context of both single and mixed infections,
and how this in turn affects transmission and
disease severity.
The underlying process of bursting infected
RBCs and invasion of uninfected RBCs is com-
mon to blood-phase malaria across animal taxa.
The methods we introduce will consequently be
generally applicable. The strength of the mouse
data we have used is the finely resolved measures
of uninfected and infected red blood cells. We are
unaware of any experimental time series in hu-
man patients in which these parameters were
directly measured, but our analyses suggest that
future longitudinal studies of individual patients
that undertake the simple assays required to di-
rectly assess RBC densities in addition to parasite
densities will lead to considerable insights into
the factors regulating human malaria.
References and Notes
1. R. Carter, D. Walliker, Ann. Trop. Med. Parasitol. 69,
187 (1975).
2. L. Molineaux, M. Träuble, W. E. Collins, G. M. Jeffery,
K. Dietz, Trans. R. Soc. Trop. Med. Hyg. 96, 205 (2002).
3. K. Dietz, G. Raddatz, L. Molineaux, Am. J. Trop. Med. Hyg.
75 (suppl.), 46 (2006).
4. D. T. Haydon, L. Matthews, R. Timms, N. Colegrave,
Proc. R. Soc. B 270, 289 (2003).
5. R. M. Ribeiro et al., J. Virol. 84, 6096 (2010).
6. O. N. Bjørnstad, B. Finkenstädt, B. T. Grenfell,
Ecol. Monogr. 72, 169 (2002).
7. R. M. Anderson, R. M. May, Infectious Diseases of
Humans (Oxford Univ. Press, Oxford, 1991).
8. M. A. Nowak, R. M. May, Virus Dynamics: Mathematical
Principles of Immunology and Virology (Oxford Univ.
Press, Oxford, 2000).
9. M. M. Stevenson, E. M. Riley, Nat. Rev. Immunol. 4,
169 (2004).
10. M. Walther et al., J. Immunol. 177, 5736 (2006).
11. M. R. Miller, L. Råberg, A. F. Read, N. J. Savill,
PLOS Comput. Biol. 6, e1000946 (2010).
12. N. Mideo et al., Am. Nat. 172, E214 (2008).
13. R. Antia, A. Yates, J. C. de Roode, Proc. R. Soc. B 275,
1449 (2008).
14. B. F. Kochin, A. J. Yates, J. C. de Roode, R. Antia,
PLoS ONE 5, e10444 (2010).
15. A. Handel, I. M. Longini Jr., R. Antia, J. R. Soc. Interface
7, 35 (2010).
16. C. L. Ball, M. A. Gilchrist, D. Coombs, Bull. Math. Biol.
69, 2361 (2007).
17. A. S. Perelson, Nat. Rev. Immunol. 2, 28 (2002).
18. R. A. Saenz et al., J. Virol. 84, 3974 (2010).
19. K. A. Lythgoe, L. J. Morrison, A. F. Read, J. D. Barry,
Proc. Natl. Acad. Sci. U.S.A. 104, 8095 (2007).
20. L. Molineaux, K. Dietz, Parassitologia 41, 221
(1999).
21. R. Killick-Kendrick, W. Peters, Eds., Rodent Malaria
(Academic Press, London, 1978).
22. V. C. Bar clay et al., Proc.R.Soc.B275, 1171
(2008).
23. S. Huijben, thesis, University of Edinburgh (2010).
24. G. H. Long, B. H. K. Chan, J. E. Allen, A. F. Read,
A. L. Graham, BMC Evol. Biol. 8, 128 (2008).
25. See supporting material on Science Online.
26. P. G. McQueen, F. E. McKenzie, Proc. Natl. Acad.
Sci. U.S.A. 101, 9161 (2004).
27. B. Hellriegel, Proc. R. Soc. B 250, 249 (1992).
28. C. Hetzel, R. M. Anderson, Parasitology 113,25
(1996).
29. W. Jarra, K. N. Brown, Parasitology 99, 157 (1989).
30. A. A. Lamikanra et al., Blood 110, 18 (2007).
31. S. S. Pilyugin, R. Antia, Bull. Math. Biol. 62, 869
(2000).
32. R. Antia, J. C. Koella, J. Theor. Biol. 168, 141 (1994).
33. J. R. Glynn, D. J. Bradley, Parasitology 110, 7 (1995).
34. M. S. Russell et al., J. Immunol. 179, 211 (2007).
35. D. L. Chao, M. P. Davenport, S. Forrest, A. S. Perelson,
Immunol. Cell Biol. 82, 55 (2004).
36. R. Stephens, J. Langhorne, PLoS Pathog. 6, e1001208
(2010).
37. C. Othoro et al., J. Infect. Dis. 179, 279 (1999).
38. F. P. Mockenhaupt et al., Blood 104, 2003 (2004).
39. S. Wambua, J. Mwacharo, S. Uyoga, A. Macharia,
T. N. Williams, Br. J. Haematol. 133, 206 (2006).
40. K. Baer, C. Klotz, S. H. Kappe, T. Schnieder, U. Frevert,
PLoS Pathog. 3, e171 (2007).
41. N. J. Savill, W. Chadwick, S. E. Reece, PLOS Comput. Biol.
5, e1000416 (2009).
42. Z. Su, A. Fortin, P. Gros, M. M. Stevenson, J. Infect. Dis.
186, 1321 (2002).
43. G. H. Long, B. H. K. Chan, J. E. Allen, A. F. Read,
A. L. Graham, Parasitology 133, 673 (2006).
44. K.-H. Chang, M. Tam, M. M. Stevenson, J. Infect. Dis.
189, 735 (2004).
Acknowledgments: Our empirical work was funded by the
Wellcome Trust (A.F.R., V.C.B., G.H.L.), the Darwin
Trust of the University of Edinburgh (S.H.), and the UK
Biotechnology and Biological Sciences Research Council
(A.L.G., G.H.L.), and the theoretical work by the Bill
and Melinda Gates Foundation (C.J.E.M., B.T.G., O.N.B.),
the RAPIDD program of the Science and Technology
Directorate (B.T.G., A.L.G., A.F.R.), and National Institute
of General Medical Sciences grant R01GM089932
(B.G., O.N.B., A.F.R.). We thank N. Mideo and P. Klepac
for extensive discussion. All authors discussed the
results and implications and commented on the
manuscript at all stages. C.J.E.M. and O.N.B. developed
the statistical approach; A.F.R., V.B., and S.H. designed
and performed the dose-dependent and CD4
+
Tcell–
depleted mice experiments; A.L.G. and G.H.L. designed
and performed the innate immunity experiments.
The authors declare no competing interests.
Supporting Online Material
www.sciencemag.org/cgi/content/full/333/6045/984/DC1
Materials and Methods
SOM Text
Figs. S1 to S12
Table S1
References
21 February 2011; accepted 22 June 2011
10.1126/science.1204588
A Large and Persistent Carbon Sink
in the World’s Forests
Yude Pan,
1
*Richard A. Birdsey,
1
Jingyun Fang,
2,3
Richard Houghton,
4
Pekka E. Kauppi,
5
Werner A. Kurz,
6
Oliver L. Phillips,
7
Anatoly Shvidenko,
8
Simon L. Lewis,
7
Josep G. Canadell,
9
Philippe Ciais,
10
Robert B. Jackson,
11
Stephen W. Pacala,
12
A. David McGuire,
13
Shilong Piao,
2
Aapo Rautiainen,
5
Stephen Sitch,
7
Daniel Hayes
14
The terrestrial carbon sink has been large in recent decades, but its size and location remain
uncertain. Using forest inventory data and long-term ecosystem carbon studies, we estimate a
total forest sink of 2.4 T0.4 petagrams of carbon per year (Pg C year
–1
) globally for 1990 to 2007.
We also estimate a source of 1.3 T0.7 Pg C year
–1
from tropical land-use change, consisting of a
gross tropical deforestation emission of 2.9 T0.5 Pg C year
–1
partially compensated by a carbon
sink in tropical forest regrowth of 1.6 T0.5 Pg C year
–1
. Together, the fluxes comprise a net global
forest sink of 1.1 T0.8 Pg C year
–1
, with tropical estimates having the largest uncertainties. Our total
forest sink estimate is equivalent in magnitude to the terrestrial sink deduced from fossil fuel
emissions and land-use change sources minus ocean and atmospheric sinks.
Forests have an important role in the global
carbon cycle and are valued globally for the
services they provide to society. International
negotiations to limit greenhouse gases require
an understanding of the current and potential
future role of forest C emissions and sequestra-
tion in both managed and unmanaged forests.
Estimates by the Intergovernmental Panel on Cli-
mate Change (IPCC) show that the net uptake by
terrestrial ecosystems ranges from less than 1.0
to as much as 2.6 Pg C year
–1
for the 1990s (1).
More recent global C analyses have estimated a
terrestrialCsinkintherangeof2.0to3.4PgC
year
–1
on the basis of atmospheric CO
2
obser-
vations and inverse modeling, as well as land
observations (2–4). Because of this uncertainty
and the possible change in magnitude over time,
constraining these estimates is critically impor-
tant to support future climate mitigation actions.
1
U.S. Department of Agriculture Forest Service, Newtown
Square, PA 19073, USA.
2
Key Laboratory for Earth Surface Pro-
cesses, Ministry of Education, Peking University, Beijing, 100871
China.
3
State Key Laboratory of Vegetation and Environmental
Change, Institute of Botany, Chinese Academy of Sciences,
Beijing, 100093 China.
4
Woods Hole Research Center, Falmouth,
MA 02543, USA.
5
University of Helsinki, Helsinki, Finland.
6
Natural
Resources Canada, Canadian Forest Service, Victoria, BC, V8Z
1M5, Canada.
7
School of Geography, University of Leeds, LS2
9JT, UK.
8
International Institute for Applied Systems Analysis,
Laxenburg, Austria.
9
Global Carbon Project, Commonwealth Sci-
entific and Industrial Research Organization Marine and Atmo-
spheric Research, Canberra, Australia.
10
Laboratoire des Sciences
du Climat et de l’Environnement CEA-UVSQ-CNRS, Gif sur Yvette,
France.
11
Duke University, Durham, NC 27708, USA.
12
Prince-
ton University, Princeton, NJ 08544, USA.
13
U.S. Geological
Survey, Alaska Cooperative Fish and Wildlife Research Unit,
University of Alaska, Fairbanks, AK 99775, USA.
14
Oak Ridge
National Laboratory, Oak Ridge, TN 37831, USA.
*To whom correspondence should be addressed. E-mail:
ypan@fs.fed.us
19 AUGUST 2011 VOL 333 SCIENCE www.sciencemag.org988
RESEARCH ARTICLES
on August 18, 2011www.sciencemag.orgDownloaded from

Here, we present bottom-up estimates of C
stocks and fluxes for the world’s forests based on
recent inventory data and long-term field obser-
vations coupled to statistical or process models
(table S1). We advanced our analyses by including
comprehensive C pools of the forest sector (dead
wood, harvested wood products, living biomass,
litter, and soil) and report past trends and changes
in C stocks across countries, regions, and conti-
nents representing boreal, temperate, and tropical
forests (5,6). To gain full knowledge of the trop-
ical C balance, we subdivided tropical forests in-
to intact and regrowth forests (Table 1). The latter
is an overlooked category, and its C uptake is
usually not reported but is implicit in the tropical
land-use change emission estimates. Although
deforestation, reforestation, afforestation and the
carbon outcomes of various management prac-
tices are included in the assessments of boreal
and temperate forest C sink estimates, we sepa-
rately estimated three major fluxes in the tropics:
C uptake by intact forests, losses from deforesta-
tion, and C uptake of forest regrowth after an-
thropogenic disturbances. The area of global
forests used as a basis for estimating C stocks and
fluxes is 3.9 billion ha, representing 95% of the
world’sforests(7) (table S2).
Global forest C stocks and changes. The
current C stock in the world’s forests is estimated
to be 861 T66 Pg C, with 383 T30 Pg C (44%) in
soil (to 1-m depth), 363 T28 Pg C (42%) in live
biomass (above and below ground), 73 T6PgC
(8%) in deadwood, and 43 T3 Pg C (5%) in litter
(table S3). Geographically, 471 T93 Pg C (55%)
is stored in tropical forests, 272 T23 Pg C (32%)
in boreal, and 119 T6PgC(14%)intemperate
forests. The C stock density in tropical and boreal
forests is comparable (242 versus 239 Mg C ha
–1
),
whereas the density in temperate forests is ~60%
of the other two biomes (155 Mg C ha
–1
).
Although tropical and boreal forests store the
most carbon, there is a fundamental difference in
their carbon structures: Tropical forests have 56%
of carbon stored in biomass and 32% in soil,
whereas boreal forests have only 20% in biomass
and 60% in soil.
The average annual change in the C stock of
established forests (Table 1) indicates a large
uptake of 2.5 T0.4PgCyear
–1
for 1990 to 1999
and a similar uptake of 2.3 T0.5PgCyear
–1
for
2000 to 2007. Adding the C uptake in tropical
regrowth forests to those values indicates a
persistent global gross forest C sink of 4.0 T0.7
Pg C year
–1
over the two periods (Tables 1 and 2).
Despite the consistency of the global C sink since
1990, our analysis revealed important regional
and temporal differences in sink sizes. The C sink
in temperate forests increased by 17% in 2000 to
2007 compared with 1990 to 1999, in contrast to
C uptake in intact tropical forests, which de-
creased by 23% (but nonsignificantly). Boreal
forests, on average, showed little difference be-
tween the two time periods (Fig. 1). Subtract-
ing C emission losses from tropical deforestation
and degradation, the global net forest C sink
was 1.0 T0.8 and 1.2 T0.9 Pg C year
–1
for
1990 to 1999 and 2000 to 2007, respectively
(Table 1).
Forest carbon sinks by regions, biomes, and
pools. Boreal forests (1135 Mha) had a consistent
average sink of 0.5 T0.1PgCyear
–1
for two dec-
ades (Table 2, 20 and 22% of the global C sinks
in established forests). However, the overall sta-
bility of the boreal forest C sink is the net result
of contrasting carbon dynamics in different boreal
countries and regions associated with natural dis-
turbances and forest management. Asian Russia
had the largest boreal sink, but that sink showed
no overall change, even with increased emissions
from wildfire disturbances (8). In contrast, there
was a notable sink increase of 35% in European
Russia (Fig. 1) attributed to several factors: in-
creased areas of forests after agricultural aban-
donment, reduced harvesting, and changes of
forest age structure to more productive stages,
particularly for the deciduous forests (8). In con-
trast to the large increase of biomass sinks in
European Russia and northern Europe (8,9), the
biomass C sink in Canadian managed forests was
reduced by half between the two periods, mostly
due to the biomass loss from intensified wildfires
and insect outbreaks (10,11). A net loss of soil C
in northern Europe was attributed to shifts of
forest to nonforest in some areas. Overall, the
relatively stable boreal C sink is the sum of a net
reduction in Canadian biomass sink offset by
increased biomass sink in all other boreal regions,
and a balance between decreased litter and soil C
sinks in northern Eurasia and a region-wide in-
crease in the accumulation of dead wood (Table 2).
Temperate forests (767 Mha) contributed 0.7 T
0.1 and 0.8 T0.1 Pg C year
–1
(27 and 34%) to
the global C sinks in established forests for two
decades (Table 2). The primary reasons for the
increased C sink in temperate forests are the
increasing density of biomass and a substantial
increase in forest area (12,13). The U.S. forest
C sink increased by 33% from the 1990s to
2000s, caused by increasing forest area; growth
of existing immature forests that are still recover-
ing from historical agriculture, grazing, harvesting
(12,14); and environmental factors such as CO
2
fertilization and N deposition (15). However, for-
ests in the western United States have shown
considerably increased mortality over the past
few decades, related to drought stress, and in-
creased mortality from insects and fires (16,17).
The European temperate forest sink was stable
between 1990 to 1999 and 2000 to 2007. There
was a large C sink in soil due to expansion of
forests in the 1990s, but this trend slowed in the
2000s (7,18). However, the increased C sink in
biomass during the second period (+17%)
helped to maintain the stability of the total C sink.
China’s forest C sink increased by 34% between
1990 to 1999 and 2000 to 2007, with the biomass
sink almost doubling (Table 2). This was caused
primarily by increasing areas of newly planted
forests, the consequence of an intensive national
afforestation/reforestation program in the past
few decades (table S2) (19).
Table 1. Global forest carbon budget (Pg C year
–1
) over two time periods. Sinks are positive values;
sources are negative values.
Carbon sink and source in biomes 1990–1999 2000–2007 1990–2007
Boreal forest 0.50 T0.08 0.50 T0.08 0.50 T0.08
Temperate forest 0.67 T0.08 0.78 T0.09 0.72 T0.08
Tropical intact forest* 1.33 T0.35 1.02 T0.47 1.19 T0.41
Total sink in global established forests†2.50 T0.36 2.30 T0.49 2.41 T0.42
Tropical regrowth forest‡1.57 T0.50 1.72 T0.54 1.64 T0.52
Tropical gross deforestation emission§ –3.03 T0.49 –2.82 T0.45 –2.94 T0.47
Tropical land-use change emission|| –1.46 T0.70 –1.10 T0.70 –1.30 T0.70
Global gross forest sink¶ 4.07 T0.62 4.02 T0.73 4.05 T0.67
Global net forest sink# 1.04 T0.79 1.20 T0.85 1.11 T0.82
Equations of global forest C fluxes
F
established forests
=F
boreal forests
+F
temperate forests
+F
tropical intact forests
(Eq. 1)
F
tropical land-use change
=F
tropical gross deforestation
+F
tropical regrowth forests
(Eq. 2)
F
gross forest sink
=F
established forests
+F
tropical regrowth forests
(Eq. 3)
F
net forest sink
=F
established forests
+F
tropical land-use change
(Eq. 4)
*Tropical intact forests: tropical forests that have not been substantially affected by direct human activities; flux accounts for the
dynamics of natural disturbance-recovery processes. †Global establishedforests: theforest remaining forest over the study periods
plus afforested land in boreal and temperate biomes, in addition to intact forest in the tropics (Eq. 1). ‡Tropical regrowth forests:
tropical forests that are recovering from past deforestation and logging. §Tropical gross deforestation: the total C emissions from
tropical deforestation and logging, notcounting the uptake of C in tropicalregrowth forests. ||Tropical land-usechange: emissions
from tropical land-use change, which is a netbalance of tropical gross deforestationemissions andC uptake in regrowth forests (Eq. 2).
It may be referenced as a tropical net deforestation emission in the literature. ¶Global gross forest sink: the sum of total sinks in
global established forests and tropical regrowth forests (Eq. 3). #Global net forest sink: the net budget of global forest fluxes
(Eq. 4). It can be calculated in two ways: (i) total sink in global established forests minus tropical land-use change emission or(ii) total
global gross forest sink minus tropical gross deforestation emission.
www.sciencemag.org SCIENCE VOL 333 19 AUGUST 2011 989
RESEARCH ARTICLES
on August 18, 2011www.sciencemag.orgDownloaded from

Tropical intact forests (1392 Mha) represent
~70% of the total tropical forest area (1949 Mha)
that accounts for the largest area of global forest
biomes (~50%). We used two networks of per-
manent monitoring sites spanning intact tropical
forest across Africa (20) and South America (21)
and assumed that forest C stocks of Southeast
Asia (9% of total intact tropical forest area) are
changing at the mean rate of Africa and South
America, as we lack sufficient data in Southeast
Asia to make robust estimates. These networks
are large enough to capture the disturbance-
recovery dynamics of intact forests (6,20,22).
We estimate a sink of 1.3 T0.3 and 1.0 T0.5 Pg C
year
–1
for 1990 to 1999 and 2000 to 2007,
Table 2. Estimated annual change in C stock (Tg C year
–1
) by biomes by country or region for the time periods of 1990 to 1999 and 2000 to 2007. Estimates
include C stock changes on “forest land remaining forest land”and “new forest land”(afforested land). The uncertainty calculation refers to the supporting online
material. ND, data not available; [1], litter is included in soils.
Biome and
country/
region
1990–1999 2000–2007
Biomass
Dead
wood Litter Soil
Harvested
wood
product
Total
stock
change
Uncertainty
(T)
Stock
change
per area Biomass
Dead
wood Litter Soil
Harvested
wood
product
Total
stock
change
Uncertainty
(T)
Stock
change
per area
(Tg C year
–1
)
(Mg C ha
–1
year
–1
) (Tg C year
–1
)
(Mg C ha
–1
year
–1
)
Boreal*
Asian
Russia 61 66 63 45 19 255 64 0.39 69 97 43 42 13 264 66 0.39
European
Russia 37 10 22 36 41 146 37 0.93 84 19 35 35 26 199 50 1.21
Canada 6–24 14 6 23 26 7 0.11 –53 16 19 7 21 10 3 0.04
European
boreal†13 0 3 38 11 65 16 1.12 21 0 4 –10 13 27 7 0.45
Subtotal 117 53 103 125 94 493 76 0.45 120 132 101 74 73 499 83 0.44
Temperate*
United
States‡118 6 13 9 33 179 34 0.72 147 9 18 37 28 239 45 0.94
Europe 117 2 8 81 24 232 58 1.71 137 2 9 65 27 239 60 1.68
China 60 22 15 31 7 135 34 0.96 115 24 8 28 7 182 45 1.22
Japan 24 9 ND 19 2 54 14 2.28 23 5 ND 8 2 37 9 1.59
South
Korea 6 2 ND 5 0 14 4 2.14 12 2 ND 4 0 18 5 2.86
Australia 17 ND 10 15 8 50 13 0.33 17 ND 10 14 10 51 13 0.34
New
Zealand 1 0 0 1 5 7 2 0.91 1 0 0 1 6 9 2 1.05
Other
countries 1 ND ND ND 0 1 1 0.07 2 0 0 0 0 3 2 0.18
Subtotal 345 42 46 160 80 673 78 0.91 454 42 45 156 80 777 89 1.03
Tropical intact
Asia 125 13 2 ND 5 144 38 0.88 100 10 2 ND 6 117 30 0.90
Africa 469 48 7 ND 9 532 302 0.94 425 43 6 ND 8 482 274 0.94
Americas 573 48 9 ND 22 652 166 0.77 345 45 5 ND 23 418 386 0.53
Subtotal 1167 109 17 ND 35 1328 347 0.84 870 98 13 ND 36 1017 474 0.71
Global
subtotal§ 1630 204 166 286 209 2494 363 0.73 1444 273 158 230 189 2294 489 0.69
Tropical regrowth
Asia 498 ND [1] 27 ND 526 263 3.52 564 ND [1] 30 ND 593 297 3.53
Africa 169 ND [1] 73 ND 242 121 1.48 188 ND [1] 83 ND 271 135 1.47
Americas 694 ND [1] 113 ND 807 403 4.67 745 ND [1] 113 ND 858 429 4.56
Subtotal 1361 ND [1] 213 ND 1574 496 3.24 1497 ND [1] 226 ND 1723 539 3.19
All tropics||
Asia 623 13 2 27 5 670 266 2.14 664 10 2 30 6 711 298 2.38
Africa 638 48 7 73 9 774 325 1.06 613 43 6 83 8 753 305 1.08
Americas 1267 48 9 113 22 1458 436 1.42 1090 45 5 113 23 1276 577 1.30
Subtotal 2529 109 17 213 35 2903 605 1.40 2367 98 13 226 36 2740 718 1.38
Global
total¶ 2991 204 166 498 209 4068 615 1.04 2941 273 158 456 189 4017 728 1.04
*Carbon outcomes of forest land-use changes (deforestation, reforestation, afforestation, and management practices) are included in the estimates in boreal and temperate forests. †Estimates for
the area that includes Norway, Sweden, and Finland. ‡Estimates for the continental U.S. and a small area in southeast Alaska. §Estimates for global established forests. ||Estimates for
all tropical forests including tropical intact and regrowth forests. ¶Areas excluded from this table include interior Alaska (51 Mha in 2007), northern Canada (118 Mha in 2007), and “other wooded
land”reported to the Food and Agriculture Organization.
19 AUGUST 2011 VOL 333 SCIENCE www.sciencemag.org990
RESEARCH ARTICLES
on August 18, 2011www.sciencemag.orgDownloaded from

respectively (Table 2). An average C sink of 1.2 T
0.4PgCyear
–1
for 1990 to 2007 is approx-
imately half of the total global C sink in estab-
lished forests (2.4 T0.4PgCyear
–1
)(Table1).
When only the biomass sink is considered, about
two-thirds of the global biomass C sink in estab-
lished forests is from tropical intact forests (1.0
versus 1.5 Pg C year
–1
). The sink reduction in the
period 2000 to 2007 (–23%) was caused by
deforestation reducing intact forest area (–8%)
and a severe Amazon drought in 2005 (21), which
appeared strong enough to affect the tropics-wide
decadal C sink estimate (–15%). Except for the
Amazon drought, the recent excess of biomass C
gain (growth) over loss (death) in tropical intact for-
ests appears to result from progressively enhanced
productivity (20,21,23). Increased dead biomass
production should lead to enhanced soil C seques-
tration, but we lack data about changes in soil C
stocks for tropical intact forests, so the C sink for
tropical intact forests may be underestimated.
Tropical land-use changes have caused net
C releases in tropical regions by clearing forests
for agriculture, pasture, and timber (24), second
in magnitude to fossil fuel emissions (Table 3).
Tropical land-use change emissions are a net
balance of C fluxes consisting of gross tropical de-
forestation emissions partially compensated by C
sinks in tropical forest regrowth. They declined
from 1.5 T0.7PgCyear
–1
in the 1990s to 1.1 T0.7
Pg C year
–1
for 2000 to 2007 (Table 1) due to
reduced rates of deforestation and increased for-
est regrowth (25). The tropical land-use change
emissions were approximately equal to the total
global land-use emissions (Tables 1 and 3), be-
cause effects of land-use changes on C were
roughly balanced in extratropics (7,24,25).
Tropical deforestation produced significant
gross C emissions of 3.0 T0.5 and 2.8 T0.5 Pg
C year
–1
, respectively, for 1990 to 1999 and 2000
to 2007, ~40% of the global fossil fuel emissions.
However, these large emission numbers are usu-
ally neglected because more than one half was
offset by large C uptake in tropical regrowth for-
ests recovering from the deforestation, logging,
or abandoned agriculture.
Tropical regrowth forests (557 Mha) repre-
sent ~30% of the total tropical forest area. The
C uptake by tropical regrowth forests is usually
implicitly included in estimated net emissions
of tropical land-use changes rather than estimated
independently as a sink (24). We estimate that
0.06
0.07
0.24
0.18
0.24
0.27
0.01
0.03
0.48
0.53
0.53
0.59
1.37
1.51
0.85
0.97
0.59
0.55
0.42
0.65
0.03
0.07
0.24
0.23 0.20
0.15
0.26
0.26
0.18
0.14
0.06
0.06
0.12
0.14
0.81
0.86
Regions of the World
Other
No Data/Other Countries
Tropical
Asia
Africa
Americas
Temperate
Continental US & S. Alaska
Europe
China
Japan/Korea
Australia/NZ
Boreal
Canada
N. Europe
Asian Russia
European Russia
Tropi ca l Regr owt h
Carbon Flux 2000-2007
Forest Carbon Flux
2000-2007
Tropi ca l Regr owt h
Carbon Flux 1990-1999
Forest Carbon Flux
1990-1999
Tropical Gross Deforestation
C Emissions 1990-1999
Tropical Gross Deforestation
C Emissions 2000-2007
Fig. 1. Carbon sinks and sources (Pg C year
–1
)intheworld’s forests. Colored
bars in the down-facing direction represent C sinks, whereas bars in the
upward-facing direction represent C sources. Light and dark purple, global
established forests (boreal, temperate, and intact tropical forests); light and
dark green, tropical regrowth forests after anthropogenic disturbances; and
light and dark brown, tropical gross deforestation emissions.
Table 3. The global carbon budget for two time periods (Pg C year
−1
). There are different arrangements to
account for elements of the global C budget (see also table S6). Here, the accounting was based on global C
sources and sinks. The terrestrial sink was the residual derived from constraints of two major anthropogenic
sourcesandthesinksintheatmosphereandoceans.WeusedtheCsinkinglobalestablishedforestsasa
proxy for the terrestrial sink.
Sources and sinks 1990–1999 2000–2007
Sources (C emissions)
Fossil fuel and cement* 6.5 T0.4 7.6 T0.4
Land-use change†1.5 T0.7 1.1 T0.7
Total sources 8.0 T0.8 8.7 T0.8
Sinks (C uptake)
Atmosphere†3.2 T0.1 4.1 T0.1
Ocean‡2.2 T0.4 2.3 T0.4
Terrestrial (established forests)§ 2.5 T0.4 2.3 T0.5
Total sinks 7.9 T0.6 8.7 T0.7
Global residuals|| 0.1 T1.0 0.0 T1.0
*See (2). †See (4,7,25). The global land-use change emission is approximately equal to the tropical land-use change emission,
because the net carbon balance of land-use changes in temperate and boreal regions is neutral (24,38). ‡See (4). §Estimates
of C sinks in the global established forests (that are outside the areas of tropical land-use changes) from this study. Note that the
carbon sink in tropical regrowth forests is excluded because it is included in the term of land-use change emission (see above and
Table 1). ||Global C residualsare close to zerowhen averaged over a decade. Uncertainties in the globalresiduals indicateeither a
land sink or source in the 212 Mha of forest not included here,on nonforest land, or systematic error in other source (overestimate) or
sink (underestimate) terms, or both.
www.sciencemag.org SCIENCE VOL 333 19 AUGUST 2011 991
RESEARCH ARTICLES
on August 18, 2011www.sciencemag.orgDownloaded from

the C sink by tropical regrowth forests was 1.6 T
0.5 and 1.7 T0.5 Pg C year
–1
, respectively, for
1990 to 1999 and 2000 to 2007. Our results in-
dicate that tropical regrowth forests were stronger
C sinks than the intact forests due to rapid bio-
mass accumulation under succession, but these
estimates are poorly constrained because of sparse
data (table S4) (6). Although distinguishing a C
sink in tropical regrowth forests does not affect
the estimated net emissions from tropical land-
use changes, an explicit estimate of this compo-
nent facilitates evaluating the complete C sink
capacity of all tropical and global forests.
When all tropical forests, both intact and
regrowth, are combined, the tropical sinks sum
to 2.9 T0.6 and 2.7 T0.7PgCyear
–1
over the
two periods (Table 1), and on average account for
~70% of the gross C sink in the world forests
(~4.0PgCyear
–1
). However, with equally
significant gross emissions from tropical de-
forestation (Table 1), tropical forests were nearly
carbon-neutral. In sum, the tropics have the world’s
largest forest area, the most intense contemporary
land-use change, and the highest C uptake, but
also the greatest uncertainty, showing that invest-
ment in better understanding carbon cycling in
the tropics should be a high priority in the future.
Deadwood, litter, soil, and harvested wood
products together accounted for 35% of the global
sink and 60% of the global forest C stock, showing
the importance of including these components
(Table 2 and table S3). Compared with biomass,
estimates of these terrestrial carbon pools are gen-
erally less certain because of insufficient data.
For deadwood, there was a large sink increase in
boreal forests over the past decade, caused by the
recent increase in natural disturbances in Siberia
and Canada. Increased deadwood carbon thus
makes a major (27%) but possibly transient con-
tribution to the total C sink in the boreal zone.
Changes in litter C accounted for a relatively
small and stable portion of the global forest C
sink. However, litter C accumulation contributed
20% of the total C sink in boreal forests and, like
deadwood, is vulnerable to wildfire disturbances.
Changes in soil C stocks accounted for more
than 10% of the total sink in the world’s forests,
largely driven by land-use changes. We may un-
derestimate global soil C stocks and fluxes be-
cause the standard 1-m soil depth excludes some
deep organic soils in boreal and tropical peat for-
ests (26–28). We estimate the net C change in har-
vested wood products (HWP), including wood in
use and disposed in landfills, as described in the
IPCC guidelines (29), attributing changes in stock
to the region where the wood was harvested. Car-
bon sequestration in HWP accounted for ~8% of
the total sink in established forests. This sink re-
mained stable for temperate and tropical regions
but declined dramatically in boreal regions because
of reduced harvest in Russia in the past decade.
Data gaps, uncertainty, and suggested im-
provements in global forest monitoring. We e s -
timated uncertainties based on a combination
of quantitative methods and expert opinions (6).
There are critical data gaps that affected both the
results presented here and our ability to report
and verify changes in forest C stocks in the fu-
ture. Data are substantially lacking for areas of
the boreal forest in North America, including
Alaska (51 Mha) and Canadian unmanaged for-
ests (118 Mha) (table S5). The forests in these
regions could be a small C source or sink, based
on the estimate of Canadian managed forests (10)
and modeling studies in Alaska (30). There is also
a lack of measurement data of soil C flux in trop-
ical intact forests, which may cause uncertainty
of 10 to 20% of the estimated total C sink in these
forest areas. In addition, there is a large uncertainty
associated with the estimate of C stocks and fluxes
in tropical Asia, due to the absence of long-term
field measurements, and a notable lack of data
about regrowth rates of tropical forests worldwide.
Prioritized recommendations for improve-
ments in regional forest inventories to assess C
density, uptake, and emissions for global-scale
aggregation include the following: (i) Land mon-
itoring should be greatly expanded in the tropics
and in unsampled regions of northern boreal
forests. (ii) Globally consistent remote sensing of
land-cover change and forest-area is required to
combine the strengths of two observation sys-
tems: solid ground truth of forest C densities
from inventories and reliable forest areas from
remote sensing. (iii) Improved methods and greater
sampling intensity are needed to estimate non-
living C pools, including soil, litter, and dead wood.
(iv) Better data are required in most regions for
estimating lateral C transfers in harvested wood
products and rivers.
Forest carbon in the global context. The new
C sink estimates from world’s forests can con-
tribute to the much needed detection and attri-
bution that is required in the context of the global
carbon budget (2,4,25). Our results suggest that,
within the limits of reported uncertainty, the en-
tire terrestrial C sink is accounted for by C uptake
of global established forests (Table 3), as the
balanced global budget yields near-zero residuals
with T1.0 Pg C year
–1
uncertainty for both 1990
to 1999 and 2000 to 2007 (Table 3). Consequent-
ly, our results imply that nonforest ecosystems
are collectively neither a major (>1 Pg) C sink
nor a major source over the two time periods that
we monitored. Because the tropical gross de-
forestation emission is mostly compensated by
the C uptakes in both tropical intact and regrowth
forests (Fig. 1 and Table 1), the net global forest
Csink(1.1T0.8PgCyear
–1
) resides mainly in
the temperate and boreal forests, consistent with
previous estimates (31,32). Notably, the total
gross C uptake by the world’s established and
tropical regrowth forests is 4.0 Pg C year
–1
,
which is equivalent to half of the fossil fuel C
emissions in 2009 (4). Over the period that we
studied (1990 to 2007), the cumulative C sink
into the world’s established forests was ~43 Pg C
and 73 Pg C for the established plus regrowing
forests; the latter equivalent to 60% of cumula-
tive fossil emissions in the period (i.e., 126 Pg C).
Clearly, forests play a critical role in the Earth’s
terrestrial C sinks and exert strong control on the
evolution of atmospheric CO
2
.
Drivers and outlook of forest carbon sink.
The mechanisms affecting the current C sink in
global forests are diverse, and their dynamics will
determine its future longevity. The C balance of
boreal forests is driven by changes in harvest
patterns, regrowth over abandoned farmlands,
and increasing disturbance regimes. The C balance
of temperate forests is primarily driven by forest
management, through low harvest rates (Europe)
(33), recovery from past harvesting and agricul-
tural abandonment (U.S.) (34), and large-scale
afforestation (China) (19). For tropical forests,
deforestation and forest degradation are dom-
inant causes of C emissions, with regrowth and
an increase in biomass in intact forests being the
main sinks balancing the emissions (23,24).
Changes in climate and atmospheric drivers
(CO
2
, N-deposition, ozone, diffuse light) affect the
C balance of forests, but it is difficult to separate
their impacts from other factors using ground
observations. For Europe, the U.S., China, and
the tropics, evidence from biogeochemical pro-
cess models suggests that climate change, in-
creasing atmospheric CO
2
, and N deposition are,
at different levels, important factors driving the
long-term C sink (15,18,20,23,34). Drought
in all regions and warmer winters in boreal re-
gions reduce the forest sink through suppressed
gross primary production, increased tree mortal-
ity, increased fires, and increased insect damage
(8,10,18,21,30,35,36).
Our estimates suggest that currently the glob-
al established forests, which are outside the areas
of tropical land-use changes, alone can account
for the terrestrial C sink (~2.4 Pg C year
–1
). The
tropics are the dominant terms in the exchange
of CO
2
between the land and the atmosphere. A
large amount of atmospheric CO
2
has been se-
questrated by the natural system of forested lands
(~4.0 Pg C year
–1
), but the benefit is substantially
offset by the C losses from tropical deforestation
(~2.9 Pg C year
–1
). This result highlights the po-
tential for the United Nations Reducing Emissions
from Deforestation and Degradation program to
lessen the risk of climate change. However, an
important caveat is that adding geological carbon
from fossil fuels into the contemporary carbon
cycle and then relying on biospheric sequestra-
tion is not without risk, because such sequestra-
tion is reversible from either climate changes, direct
human actions, or a combination of both.
Nonetheless, C sinks in almost all forests
across the world (Fig. 1) may suggest overall fa-
vorable conditions for increasing stocks in forests
and wood products. Our analysis also suggests
that there are extensive areas of relatively young
forests with potential to continue sequestering
C in the future in the absence of accelerated
natural disturbance, climate variability, and land-
use change. As a result of the large C stocks in
both boreal forest soils and tropical forest bio-
mass, warming in the boreal zone, deforestation,
19 AUGUST 2011 VOL 333 SCIENCE www.sciencemag.org
992
RESEARCH ARTICLES
on August 18, 2011www.sciencemag.orgDownloaded from

and occasional extreme drought, coincident with
fires in the tropics, represent the greatest risks
to the continued large C sink in the world’sfor-
ests (21,24,30,37). A better understanding of
the role of forests in biosphere C fluxes and mech-
anisms responsible for forest C changes is critical
for projecting future atmospheric CO
2
growth
and guiding the design and implementation of
mitigation policies.
Reference and Notes
1. G. J. Nabuurs et al., in Climate Change 2007: Mitigation,
B.Metz,O.R.Davidson,P.R.Bosch,R.Dave,L.A.Meyer,Eds.
(Cambridge Univ. Press, Cambridge, 2007), pp. 542–584.
2. J. G. Canadell et al., Proc. Natl. Acad. Sci. U.S.A. 104,
18866 (2007).
3. S. Khatiwala, F. Primeau, T. Hall, Nature 462, 346 (2009).
4. C. Le Quéré et al., Nat. Geosci. 2, 831 (2009).
5. R. K. Dixon et al., Science 263, 185 (1994).
6. Details of data sources, accounting, and estimation
methods used for each country, region, and C component
are provided in the supporting online material.
7. Food and Agriculture Organization, Global Forest
Resources Assessment 2010 (Food and Agriculture
Organization, Rome, 2010), forestry paper 163.
8. A. Z. Shvidenko, D. G. Schepaschenko, S. Nilsson, in
Basic Problems of Transition to Sustainable Forest
Management in Russia, V. A. Sokolov, A. Z. Shvidenko,
O. P. Vtorina, Eds. (Russian Academy of Sciences,
Krasnoyarsk, Russia, 2007), pp. 5–35.
9. P. E. Kauppi et al., For. Ecol. Manage. 259, 1239 (2010).
10. W. A. Kurz, G. Stinson, G. J. Rampley, C. C. Dymond,
E. T. Neilson, Proc. Natl. Acad. Sci. U.S.A. 105, 1551 (2008).
11. G. Stinson et al., Glob. Change Biol. 17, 2227 (2011).
12. R. Birdsey, K. Pregitzer, A. Lucier, J. Environ. Qual. 35,
1461 (2006).
13. P. E. Kauppi et al., Proc. Natl. Acad. Sci. U.S.A. 103,
17574 (2006).
14. Y. Pan et al., Biogeosciences 8, 715 (2011).
15. Y. Pan, R.Birdsey, J. Hom,K. McCullough, For. Ecol. Manage.
259, 151 (2009).
16. P. J. van Mantgem et al., Science 323, 521 (2009).
17. D. D. Breshears et al., Proc. Natl. Acad. Sci. U.S.A. 102,
15144 (2005).
18. P. Ciais et al., Nat. Geosci. 1, 425 (2008).
19. J. Fang, A. Chen, C. Peng, S. Zhao, L. Ci, Science 292,
2320 (2001).
20. S. L. Lewis et al., Nature 457, 1003 (2009).
21. O. L. Phillips et al., Science 323, 1344 (2009).
22. M. Gloor et al., Glob. Change Biol. 15, 2418 (2009).
23. S. L. Lewis, J. Lloyd, S. Sitch, E. T. A. Mitchard,
W. F. Laurance, Annu. Rev. Ecol. Syst. 40, 529 (2009).
24. R. A. Houghton, Annu. Rev. Earth Planet. Sci. 35, 313 (2007).
25. P. Friedlingstein et al., Nat. Geosci. 3, 811 (2010).
26. C. Tarnocai et al., Global Biogeochem. Cycles 23,
GB2023 (2009).
27. A. Hooijer et al., Biogeosciences 7, 1505 (2010).
28. S. E. Page, J. O. Rieley, C. J. Banks, Glob. Change Biol.
17, 798 (2011).
29. Intergovernmental Panel on Climate Change, IPCC
Guidelines for National Greenhouse Gas Inventories
(Institute for Global Environmental Strategies, Japan, 2006);
www.ipcc-nggip.iges.or.jp/public/2006gl/index.html.
30. A. D. McGuire et al., Ecol. Monogr. 79, 523 (2009).
31. C. L. Goodale et al., Ecol. Appl. 12, 891 (2002).
32. J. L. Sarmiento et al., Biogeosciences 7, 2351 (2010).
33. E. D. Schulze et al., Nat. Geosci. 2, 842 (2009).
34. S. W. Pacala et al., Science 292, 2316 (2001).
35. O. L. Phillips et al., Philos. Trans. R. Soc. London Ser. B
359, 381 (2004).
36. J. M. Metsaranta, W. A. Kurz, E. T. Neilson, G. Stinson,
Tellus 62B, 719 (2010).
37. M. Zhao, S. W. Running, Science 329, 940 (2010).
38. R. A. Houghton, Tellus 55B, 378 (2003).
Acknowledgments: This study is the major output of two
workshops at Peking Univ. and Princeton Univ. Y.P.,
R.A.B., and J.F. were lead authors and workshop
organizers; Y.P., R.A.B., J.F., R.H., P.E.K., W.A.K., O.L.P.,
A.S., and S.L.L. contributed primary data sets and
analyses; J.G.C., P.C., R.B.J., and S.W.P. contributed
noteworthy ideas to improve the study; A.D.M., S.P.,
A.R., S.S., and D.H. provided results of modeling or
data analysis relevant to the study; and all authors
contributed in writing, discussions, or comments. We
thank K. McCullough for helping to make the map in
Fig. 1 and C. Wayson for helping to develop a
Monte-Carlo analysis. This work was supported in part
by the U.S. Forest Service, NASA (grant 31021001), the
National Basic Research Program of China on Global
Change (2010CB50600), the Gordon and Betty Moore
Foundation, Peking Univ., and Princeton Univ. This work
is a contribution toward the Global Carbon Project’s aim
of fostering an international framework to study the
global carbon cycle.
Supporting Online Material
www.sciencemag.org/cgi/content/full/science.1201609/DC1
Materials and Methods
SOM Text
Tables S1 to S6
References
13 December 2010; accepted 29 June 2011
Published online 14 July 2011;
10.1126/science.1201609
REPORTS
Detection of Emerging Sunspot
Regions in the Solar Interior
Stathis Ilonidis,*Junwei Zhao, Alexander Kosovichev
Sunspots are regions where strong magnetic fields emerge from the solar interior and where major
eruptive events occur. These energetic events can cause power outages, interrupt telecommunication
and navigation services, and pose hazards to astronauts. We detected subsurface signatures of
emerging sunspot regions before they appeared on the solar disc. Strong acoustic travel-time anomalies
of an order of 12 to 16 seconds were detected as deep as 65,000 kilometers. These anomalies were
associated with magnetic structures that emerged with an average speed of 0.3 to 0.6 kilometer per
second and caused high peaks in the photospheric magnetic flux rate 1 to 2 days after the detection of
the anomalies. Thus, synoptic imaging of subsurface magnetic activity may allow anticipation of large
sunspot regions before they become visible, improving space weather forecast.
Understanding solar magnetism is among
the most important problems of solar phys-
ics and astrophysics (1–5). Modern theo-
ries assume that sunspot regions are generated by
a dynamo action at the bottom of the convection
zone, about 200 Mm below the photosphere. How-
ever, there is no convincing observational evidence
to support this idea, and dynamo mechanisms op-
erating in the bulk of the convection zone or even
in the near-surface shear layer have been pro-
posed as well (6,7). Investigation of emerging
magnetic flux could possibly determine the depth
of this process and set the foundations for a better
understanding of sunspots and active regions.
Active regions on the Sun produce flares and
mass eruptions that may cause power outages
on Earth, satellite failures, and interruptions of
telecommunication and navigation services. Moni-
toring solar subsurface processes and predict-
ing magnetic activity would also improve space
weather forecasts.
Time-distance helioseismology (8) is one of
the local helioseismology techniques that image
acoustic perturbations in the interior of the Sun
(9). Acoustic waves are excited by turbulent con-
vection near the surface, propagate deep inside
the Sun, and are refracted back to the surface
(Fig. 1). Time-distance helioseismology measures
travel times of acoustic waves propagating to dif-
ferent distances by computing cross-covariances
between the oscillation signals observed at pairs
of locations on the solar photosphere. Varia-
tions in acoustic travel times are caused mainly
by thermal perturbations, magnetic fields, and
flows. Previous studies of emerging sunspot re-
gions (10–14) have found difficulties in detect-
ing signals deeper than 30 Mm and before the
initial magnetic field becomes visible on the sur-
face because of the fast emergence speed and low
signal-to-noise ratio (15). Here, we present a deep-
focus time-distance measurement scheme, which
allows us to detect signals of emerging magnetic
regions in the deep solar interior (16,17).
We have used Doppler observations (18) from
Michelson Doppler Imager (MDI) (19) onboard
the Solar and Heliospheric Observatory (SOHO)
and computed travel-time maps of four emerging
flux regions and nine quiet regions. In Fig. 2, we
present the results of our analysis for Active Re-
gion (AR) 10488, which started emerging on the
solar disc at 09:30 UT, 26 October 2003, about
W. W. Hansen Experimental Physics Laboratory, Stanford
University, Stanford, CA 94305–4085, USA
*To whom correspondence should be addressed. E-mail:
ilonidis@sun.stanford.edu
www.sciencemag.org SCIENCE VOL 333 19 AUGUST 2011 993
on August 18, 2011www.sciencemag.orgDownloaded from