Increasing River Discharge to
the Arctic Ocean
Bruce J. Peterson,
1
* Robert M. Holmes,
1
James W. McClelland,
1
Charles J. Vo¨ro¨smarty,
2
Richard B. Lammers,
2
Alexander I. Shiklomanov,
2
Igor A. Shiklomanov,
3
Stefan Rahmstorf
4
Synthesis of river-monitoring data reveals that the average annual discharge of
fresh water from the six largest Eurasian rivers to the Arctic Ocean increased
by 7% from 1936 to 1999. The average annual rate of increase was 2.0 ⫾ 0.7
cubic kilometers per year. Consequently, average annual discharge from the six
rivers is now about 128 cubic kilometers per year greater than it was when
routine measurements of discharge began. Discharge was correlated with
changes in both the North Atlantic Oscillation and global mean surface air
temperature. The observed large-scale change in freshwater flux has potentially
important implications for ocean circulation and climate.
The Arctic is expected to be disproportion-
ately affected by global warming, and the
Arctic in turn may exert strong feedback on
global climate (1). Many of the linkages be-
tween the arctic system and global climate
involve the hydrologic cycle, including atmo-
spheric moisture transport from lower to
higher latitudes (2). This transport of mois-
ture is predicted to increase with warming
(1); both theoretical arguments and models
suggest that net high-latitude precipitation
increases in proportion to increases in mean
hemispheric temperature (3, 4 ). At the same
time, freshening of the Arctic Ocean is ex-
pected to reduce North Atlantic Deep Water
(NADW) formation and Atlantic thermoha-
line circulation (THC) (3–6 ).
Global surface air temperature (SAT) has
increased 0.6° ⫾ 0.2°C over the past century
(1). Thus, one might expect that net moisture
transport into the pan-arctic drainage basin
(Fig. 1) is already increasing. Are such
changes detectable in historical data sets?
Arctic river discharge is particularly useful
for addressing this question, because it pro-
vides an integrative measure of the continen-
tal water balance. The pan-arctic drainage
covers 22.4 ⫻ 10
6
km
2
(7 ), an area ⬃1.5
times that of the Arctic Ocean. Consequently,
the Arctic Ocean is the most land-dominated
of ocean basins. Net precipitation ( precipita-
tion minus evaporation) over the pan-arctic
drainage is recorded in river discharge, the
majority of which is routed to the Arctic
Ocean through a handful of very large and
well-monitored rivers (8).
Evidence of increasing arctic river dis-
charge has been reported in several recent
publications, with changes most evident
during the low-flow period from November
through April (7, 9–11). However, analyses
of trends have emphasized different time
periods and areas of the Arctic (12). Thus,
it has been difficult to generalize about
temporal trends in discharge to the Arctic
Ocean. In particular, high interannual vari-
ation makes it difficult to discern trends
over shorter time periods. Discharge
records for downstream stations on major
North American arctic rivers extend back
only a few decades, whereas gauging of
major Eurasian arctic rivers generally be-
gan in the 1930s. The longer records pro-
vide the best opportunity for detecting
change. Here we identify long-term trends
in discharge from major Eurasian rivers to
the Arctic Ocean and evaluate their possi-
ble links to climate variability.
Total river inflow to the Arctic Ocean is
dominated by contributions from Eurasia
(Fig. 1). The six largest Eurasian arctic
rivers (Yenisey, Lena, Ob’, Pechora,
Kolyma, and Severnaya Dvina) drain about
two-thirds of the Eurasian arctic landmass
and include three of the largest rivers on
Earth. Over the period of record (from 1936
to 1999), aggregate annual discharge from
the six largest Eurasian arctic rivers shows
a significant positive trend (Fig. 2). The
average annual rate of increase was 2.0 ⫾
0.7 km
3
/year. Thus, average annual dis-
charge is now about 128 km
3
/year (0.004
sverdrup) greater than it was when routine
measurements of discharge from the major
Eurasian arctic rivers began in the 1930s.
This amounts to an approximate 7% in-
crease in discharge. Although long-term
trends in discharge are difficult to detect in
the individual rivers (fig. S1), the additive
effect of discharge increases in these rivers
results in a strong signal of change at the
scale of the Eurasian Arctic as a whole
(13).
Correspondence between Eurasian river
discharge and the North Atlantic Oscillation
(NAO) suggests that the rivers are responding
to changes in large-scale hemispheric climate
patterns (Fig. 3). This relationship is consis-
tent with the report by Serreze et al. (14 ) that
cyclone abundance and intensity in the North
Atlantic have increased in parallel with the
ramping-up of the NAO over the past few
decades. Furthermore, Dickson et al. (15)
1
The Ecosystems Center, Marine Biological Laborato-
ry, Woods Hole, MA 02543, USA.
2
Water Systems
Analysis Group, Institute for the Study of Earth,
Oceans, and Space, University of New Hampshire,
Durham, NH 03824, USA.
3
State Hydrological Insti-
tute, St. Petersburg, Russia 199053.
4
Potsdam Insti-
tute for Climate Impact Research, Box 601203, 14412
Potsdam, Germany.
*To whom correspondence should be addressed. E-
mail: peterson@mbl.edu
Fig. 1. Map of the
pan-arctic watershed
showing catchments
and average annual
discharge of the ma-
jor Eurasian rivers
that contribute water
to the Arctic Ocean.
Generalized NADW
source locations in
the Greenland-Ice-
land-Norwegian and
Labrador seas are in-
dicated by red dots.
R EPORTS
www.sciencemag.org SCIENCE VOL 298 13 DECEMBER 2002 2171
provide a spatial analysis of precipitation
that indicates that high-NAO years are
characterized by positive precipitation
anomalies that are greatest in Scandinavia
but extend across Siberia to the Lena River
watershed. Whether the recent NAO trend
is a response to global warming is unknown
(15, 16 ).
Increases in discharge also correspond
to increases in global, pan-arctic, and Eur-
asian arctic SATs (14 ). Over the period of
the discharge record, global SAT increased
by 0.4°C, pan-arctic SAT increased by
0.6°C, and Eurasian arctic SAT increased
by 0.7°C(17 ). Although temperature in-
creases at these different scales are all in-
terrelated, the most appropriate relationship
for considering increased moisture trans-
port from lower to higher latitudes is be-
tween global SAT and river discharge (Fig.
3). Such a linkage has been predicted by
several general circulation models (GCMs)
and arctic hydrologic models (18–20).
Regression analysis of discharge against
global SAT using annual data from 1936 to
1999 identified a positive relationship (Fig.
4). The slope in Fig. 4 (0.007 sverdrup/°C)
provides an empirical estimate of a major
component of the Atlantic hydrologic sen-
sitivity parameter (HSP); the Atlantic HSP
describes the amount of additional freshwa-
ter predicted to enter the Arctic Ocean/
Atlantic Ocean north of 50°N per °Cof
global warming (21). Hydrologic sensitivi-
ty is the main control variable that deter-
mines the future response of THC. It is also
one of the largest sources of uncertainty in
predicting this response, because hydrolog-
ic sensitivity is, at the moment, poorly
constrained by observations.
GCM predictions of future changes in
global SAT are reasonably well constrained
(1). Consequently, GCM estimates of glob-
al SAT provide a tool for projecting dis-
charge trends into the future. The Intergov-
ernmental Panel on Climate Change (IPCC)
(1) projects a global SAT rise between 1.4°
and 5.8°C by 2100. Thus, at a rate of 0.007
sverdrup/°C (Fig. 4), discharge from the six
largest Eurasian arctic rivers alone would
increase by 0.01 to 0.04 sverdrup (315 to
1260 km
3
/year) by 2100. This would rep-
resent an 18 to 70% increase in Eurasian
arctic river discharge over present condi-
tions. A comparable increase in Eurasian
arctic river discharge (⬃35%) has been
predicted by the NASA Goddard Institute
for Space Studies GCM in response to a
twofold increase in atmospheric CO
2
(⬃4°C global warming) (22, 23).
Changes in river discharge of this mag-
nitude are potentially important with re-
spect to NADW formation. Increases in
discharge from other arctic rivers (24 ), net
precipitation directly over the ocean, and
meltwater from Greenland would provide
additional freshwater forcing (25). Fresh-
water sensitivity experiments with a range
of ocean and climate models predict a crit-
ical bifurcation point between 0.06 and
0.15 sverdrup of additional freshwater en-
tering the northern Atlantic, after which
NADW formation cannot be sustained (5,
6, 26–28). Although the extrapolation of
arctic river discharge presented here should
not be regarded as a prediction, it does
highlight the potential importance of arctic
river discharge as a feedback on Atlantic
THC within the 21st century. Thus, it is
particularly urgent now to increase our un-
derstanding of the coupled land, ocean, and
atmospheric components of the arctic hy-
drologic cycle.
Fig. 2. Combined an-
nual discharge from the
six largest Eurasian arctic
rivers (for the period
from 1936 to 1999).
Values are from the
R-ArcticNet database
(www.r-arcticnet.sr.unh.
edu/), updated to 1999.
Discharge data from the
Yenisey at Igarka, Lena
at Kyusyur, Ob’ at Sale-
khard, Kolyma at
Kolymskoye, Pechora at
Ust’ Tsil’ma, and Sever-
naya Dvina at Ust’ Pin-
ega were summed to
obtain the aggregate
values for Eurasia. The
trendline, slope, and P value are from a simple linear regression of time versus discharge. The data sets for
gauging stations at Igarka ( Yenisey), Kyusyur (Lena), and Salekhard (Ob’) are complete from 1936 to 1999.
Discharge values for 1982 at Ust’ Pinega (Severnaya Dvina) and 1999 at Ust’ Tsil’ma (Pechora) have been
estimated from values of surrounding years. Values for Kolymskoye (Kolyma) between 1936 and 1977 have
been derived from the next upstream gauging station at Srednekolymsk, according to the linear equation
relating this station to the one at Kolymskoye for an overlapping period of 8 years (r
2
⫽ 0.99). Sv, sverdrup.
Fig. 3. Ten-year running
averages of the Eurasian
arctic river discharge
anomaly, winter (De-
cember through March)
NAO index, and global
mean SAT for 1936 to
1999. “Anomaly” refers
to variation from the
long-term mean. NAO
data were taken from
www.cgd.ucar.edu/⬃
jhurrell/nao.html, and
temperature data are
fromwww.giss.nasa.gov/
data/update/gistemp/.
Fig. 4. Annual Eurasian
arctic river discharge
anomaly versus global
surface air temperature
(from 1936 to 1999).
R EPORTS
13 DECEMBER 2002 VOL 298 SCIENCE www.sciencemag.org2172
References and Notes
1. Intergovernmental Panel on Climate Change, Climate
Change 2001: The Scientific Basis. Contribution of
Working Group I to the Third Assessment Report of
the IPCC, J. T. Houghton et al., Eds. (Cambridge Univ.
Press, Cambridge, 2001).
2. C. Vo¨ro¨smarty et al., Eos 83, 241 (2002).
3. S. Manabe, R. J. Stouffer, J. Clim. 7, 5 (1994).
4. S. Rahmstorf, A. Ganopolski, Clim. Change 43, 353
(1999).
5. S. Rahmstorf, Nature 378, 145 (1995).
6. W. S. Broecker, Science 278, 1582 (1997).
7. R. B. Lammers, A. I. Shiklomanov, C. J. Vo¨ro¨smarty,
B. J. Peterson, J. Geophys. Res. 106, 3321 (2001).
8. The Russian Federal Service for Hydrometeorology
and Environment Monitoring (Roshydromet) moni-
tors the discharge of Russian arctic rivers. We have
compiled and digitized this database, which is avail-
able on our Web site (www.r-arcticnet.sr.unh.edu/)
and from the National Snow and Ice Data Center
(http://nsidc.org). The downstream stations from
which the data in this paper come have been moni-
tored almost continuously from 1936 to 1999, de-
spite a general decline in the arctic hydrologic mon-
itoring network that began in the mid 1980s (29).
Roshydromet considers these stations to be of pri-
mary importance because of their proximity to the
ocean. Stage height readings to the nearest centime-
ter were made daily, and cross-channel measure-
ments of discharge for rating curve calibrations were
made 25 to 30 times per year in each river. With
these frequent calibrations, estimates of daily dis-
charge from the rating curves were accurate to ⫾5%.
9. I. P. Semiletov et al.,inThe Freshwater Budget of the
Arctic Ocean, E. L. Lewis, Ed. (Kluwer Academic, Dor-
drecht, Netherlands, 2000), pp. 323–366.
10. I. A. Shiklomanov, A. I. Shiklomanov, R. B. Lammers,
B. J. Peterson, C. J. Vo¨ro¨smarty, in The Freshwater
Budget of the Arctic Ocean, E. L. Lewis, Ed. (Kluwer
Academic, Dordrecht, Netherlands, 2000), pp. 281–
296.
11. N. I. Savelieva, I. P. Semiletov, L. N. Vasilevskaya, S. P.
Pugach, Prog. Oceanogr. 47, 279 (2000).
12. T. D. Prowse, P. O. Flegg, Hydrol. Processes 14, 3185
(2000).
13. The statistical strength of the temporal trend in
combined river discharge from the Eurasian arctic is
greater than for individual rivers, because the esti-
mated slope for the sum increases with the number
of rivers but the standard error of this estimate
increases only with the square root of the number of
rivers. This leads to a more powerful test of the null
hypothesis of no trend.
14. M. C. Serreze et al., Clim. Change 46, 159 (2000).
15. R. R. Dickson et al., J. Clim. 13, 2671 (2000).
16. R. E. Moritz, C. M. Bitz, E. J. Steig, Science 297, 1497
(2002).
17. Global temperature changes were calculated using
data from the NASA Goddard Institute for Space
Studies (www.giss.nasa.gov/data/update/gistemp/),
and temperature changes for the pan-arctic and Eur-
asian arctic were calculated using the database of
New et al. (30).
18. B. Nijssen, G. M. O’Donnell, A. F. Hamlet, D. P.
Lettenmaier, Clim. Change 50, 143 (2001).
19. J. R. Miller, G. L. Russell, Geophys. Res. Lett. 27, 1183
(2000).
20. M. Stieglitz, A. Giblin, J. Hobbie, M. Williams, G. Kling,
Global Biogeochem. Cycles 14, 1123 (2000).
21. HSP is defined in (4). In addition to river runoff, the
Atlantic HSP includes meltwater from sea and conti-
nental ice and precipitation minus evaporation (P – E)
over the ocean. The difference between basing hy-
drologic sensitivity on global or hemispheric temper-
ature is small, because in greenhouse gas–induced
warming, both hemispheres warm by similar
amounts. When linking hydrologic changes to local
temperature, such as for Greenland, one needs to
take into account that local warming over Greenland
could be twice the global mean; a ⬃10% accumula-
tion increase per °C of local warming over Greenland,
as found in Greeland Ice Core Project and Greenland
Ice Sheet Project 2 ice core data, is then consistent
with the slope shown in Fig. 4.
22. J. R. Miller, G. L. Russell, J. Geophys. Res. 97, 2757
(1992).
23. S. Van Blarcum, J. R. Miller, G. L. Russell, Clim. Change
30, 7 (1995).
24. It is difficult to discern long-term trends for major
North American arctic rivers, because discharge
records from downstream gauging stations are rela-
tively short (⬃25 to 35 years). Regression analyses of
Mackenzie (at Arctic Red) and Yukon (at Pilot Station)
discharge records produces slopes of ⫹0.6 km
3
/year
and – 0.1 km
3
/year for the two rivers, respectively,
but neither is a statistically significant result. How-
ever, temperature trends over North America that
are similar to Eurasian trends over the past century
support potential arctic-wide increases in river dis-
charge (30).
25. Modeling studies predict substantial increases in
freshwater contributions from the Greenland ice
sheet (31, 32). Although arctic glaciers and sea ice are
much smaller in volume than the Greenland ice
sheet, additional freshwater could be derived from
their meltdown (33, 34). Finally, P – E over the Arctic
Ocean appears to have increased over the past cen-
tury and is expected to continue to increase in the
future (19).
26. S. Rahmstorf, Clim. Dyn. 12, 799 (1996).
27. P. U. Clark, N. G. Pisias, T. F. Stocker, A. J. Weaver,
Nature 415, 863 (2002).
28. S. Rahmstorf, Nature 419, 207 (2002).
29. A. I. Shiklomanov, R. B. Lammers, C. J. Vo¨ro¨smarty,
Eos 83, 13 (2002).
30. M. New, M. Hulme, P. Jones, J. Clim. 13, 2217 (2000).
31. P. Huybrechts, J. DeWolde, J. Clim. 12, 2169 (1999).
32. M. Wild, A. Ohmura, Ann. Glaciol. 30, 197 (2000).
33. J. A. Dowdeswell et al., Quat. Res. 48, 1 (1997).
34. K. Y. Vinnikov et al., Science 286, 1934 (1999).
35. We gratefully acknowledge the efforts of the
personnel of the Russian Federal Service for
Hydrometeorology and Environment Monitoring.
We thank A. Solow of the Woods Hole Oceano-
graphic Institute for thoughtful comments and
advice on statistical analyses and R. Alley, I. Belkin,
T. Stocker, D. Schindler, M. Stieglitz, and G. Hollo-
way for their insights and advice. Supported by
the Arctic System Science Program of NSF (grants
NSF-OPP-9524740, NSF-OPP-9818199, and
NSF-OPP-0229302) and NASA (NASA EOS grant
NAG5-6137).
Supporting Online Material
www.sciencemag.org/cgi/content/full/298/5601/2171/
DC1
Fig. S1
16 August 2002; accepted 7 November 2002
Soil Warming and Carbon-Cycle
Feedbacks to the Climate
System
J. M. Melillo,
1
* P. A. Steudler,
1
J. D. Aber,
2
K. Newkirk,
1
H. Lux,
1
F. P. Bowles,
3
C. Catricala,
1
A. Magill,
2
T. Ahrens,
1
S. Morrisseau
1
In a decade-long soil warming experiment in a mid-latitude hardwood forest,
we documented changes in soil carbon and nitrogen cycling in order to inves-
tigate the consequences of these changes for the climate system. Here we show
that whereas soil warming accelerates soil organic matter decay and carbon
dioxide fluxes to the atmosphere, this response is small and short-lived for a
mid-latitude forest, because of the limited size of the labile soil carbon pool.
We also show that warming increases the availability of mineral nitrogen to
plants. Because plant growth in many mid-latitude forests is nitrogen-limited,
warming has the potential to indirectly stimulate enough carbon storage in
plants to at least compensate for the carbon losses from soils. Our results
challenge assumptions made in some climate models that lead to projections
of large long-term releases of soil carbon in response to warming of forest
ecosystems.
The acceleration of global warming due to
terrestrial carbon-cycle feedbacks may be an
important component of future climate
change (1). Recent experiments with fully
coupled, three-dimensional carbon-climate
models suggest that carbon-cycle feedbacks
could substantially accelerate (2) or slow (3)
climate change over the 21st century. Both
the sign and magnitude of these feedbacks in
the real Earth system are still highly uncertain
because of gaps in basic understanding of
terrestrial ecosystem processes (2, 3). For
example, the potential switch of the terrestrial
biosphere from its current role as a carbon
sink (4, 5) to a carbon source is critically
dependent upon the long-term sensitivity to
global warming of the respiration of soil mi-
crobes, which is still a subject of debate
(6–8). Here we present results from a long-
term (10-year) soil warming experiment de-
signed to explore this feedback issue in an
ecosystem context.
We began our soil warming study in April
1991, in an even-aged mixed hardwood forest
stand at the Harvard Forest in central Massa-
chusetts (42.54°N, 72.18°W). Dominant tree
1
The Ecosystems Center, Marine Biological Laborato-
ry, Woods Hole, MA 02543, USA.
2
Complex Systems
Research Center, University of New Hampshire,
Durham, NH 03824, USA.
3
Research Designs, Post
Office Box 26, Woods Hole, MA 02543, USA.
*To whom correspondence should be addressed. E-
mail: jmelillo@mbl.edu
R EPORTS
www.sciencemag.org SCIENCE VOL 298 13 DECEMBER 2002 2173