Northern Hemisphere atmospheric stilling partly attributed to an increase in surface roughness

Abstract

Surface winds have declined in China, the Netherlands, the Czech Republic, the United States and Australia over the past few decades. The precise cause of the stilling is uncertain. Here, we analyse the extent and potential cause of changes in surface wind speeds over the northern mid-latitudes between 1979 and 2008, using data from 822 surface weather stations. We show that surface wind speeds have declined by 5g-15% over almost all continental areas in the northern mid-latitudes, and that strong winds have slowed faster than weak winds. In contrast, upper-air winds calculated from sea-level pressure gradients, and winds from weather reanalyses, exhibited no such trend. Changes in atmospheric circulation that are captured by reanalysis data explain 10g-50% of the surface wind slowdown. In addition, mesoscale model simulations suggest that an increase in surface roughnessg-the magnitude of which is estimated from increases in biomass and land-use change in Eurasiag-could explain between 25 and 60% of the stilling. Moreover, regions of pronounced stilling generally coincided with regions where biomass has increased over the past 30 years, supporting the role of vegetation increases in wind slowdown.

Full text

LETTERS
PUBLISHED ONLINE: 17 OCTOBER 2010 | DOI: 10.1038/NGEO979
Northern Hemisphere atmospheric stilling partly
attributed to an increase in surface roughness
Robert Vautard1*, Julien Cattiaux1, Pascal Yiou1, Jean-Noël Thépaut2and Philippe Ciais1
Surface winds have declined in China, the Netherlands, the
Czech Republic, the United States and Australia over the
past few decades1–4. The precise cause of the stilling is
uncertain. Here, we analyse the extent and potential cause of
changes in surface wind speeds over the northern mid-latitudes
between 1979 and 2008, using data from 822 surface weather
stations. We show that surface wind speeds have declined
by 5–15% over almost all continental areas in the northern
mid-latitudes, and that strong winds have slowed faster than
weak winds. In contrast, upper-air winds calculated from sea-
level pressure gradients, and winds from weather reanalyses,
exhibited no such trend. Changes in atmospheric circulation
that are captured by reanalysis data explain 10–50% of
the surface wind slowdown. In addition, mesoscale model
simulations suggest that an increase in surface roughness—the
magnitude of which is estimated from increases in biomass
and land-use change in Eurasia—could explain between 25 and
60% of the stilling. Moreover, regions of pronounced stilling
generally coincided with regions where biomass has increased
over the past 30 years, supporting the role of vegetation
increases in wind slowdown.
The decline of surface wind observed in many regions of
the world is a potential concern for wind power electricity
production5, and has been shown to be the main cause of decreasing
pan evaporation6–8. In China, a persistent decrease of monsoon
winds was observed in all seasons3,9. Stilling winds were also
evidenced over the Netherlands2, in the Czech Republic10, over
the conterminous US (refs 1,11) and most of Australia4. In
Mediterranean regions, wind trends were non-monotonic over
the past decades12. At high latitudes, the surface winds were
found to increase13,14.
Such surface wind trends can be due to (1) changes in mean
circulation15,16 and/or to the decrease of synoptic weather system
intensity, both as a consequence of climate change, (2) changes
in near-surface wind due to increasing surface roughness in the
near field of each station2,4,11 and/or in boundary layer structure,
and/or (3) instrumental or observational drifts17,18. Over China,
wind decline was attributed to a north–south warming gradient in
winter, and to sunlight dimming caused by air pollution lingering
over central areas3,9 in summer. No clear explanation was given for
the wind decline in other studied regions of interest.
The attribution of the stilling drivers requires a global investi-
gation of available surface and upper-air wind data, which has not
been conducted so far. Here we use global data sets of in situ wind
measurements (see the Methods section and Supplementary Infor-
mation). A set of 822 worldwide surface stations with continuous
wind records was selected after careful elimination of stations with
obvious breaks and large gaps. This data set covers most of the
northern mid-latitudes over the 1979–2008 period.
1LSCE/IPSL, Laboratoire CEA/CNRS/UVSQ, 91191 Gif/Yvette Cedex, France, 2ECMWF, Shinfield Park, Reading RG2 9AX, UK.
*e-mail: robert.vautard@lsce.ipsl.fr.
We found that annual mean wind speeds have declined at 73% of
surface stations over the past 30 years (Fig. 1a). In Europe, Central
Asia, Eastern Asia and in North America (Fig. 1a) the annual mean
surface wind speed has decreased on average at a rate of −0.09,
−0.16, −0.12 and −0.07 m s−1decade−1, respectively (−2.9, −5.9,
−4.2 and −1.8% per decade), that is, a decrease of about 10%
in 30 years and up to almost 20% in Central Asia, where wind
speed trends have not been studied so far. These numbers are all
statistically significant (p<0.1% for the regression coefficient).
Tropical areas are not well covered by the data set. However the
wind decline over South Asia is also about −5% per decade, and
−0.08 m s−1per decade.
Zonal means of wind trends exhibit a rather homogeneous
behaviour across latitudes, in the range of −0.06 to −0.11 to
m s−1decade−1. A higher frequency of positive trends is found
only over the ∼15 northern polar latitude stations. A few wind
series starting earlier (after 1959) further indicate that stilling
actually started at least as early as in the 1960s (Supplementary Fig.
S1). Negative trends were also found in mid to high percentiles
of the wind speed distribution (see Supplementary Table S1),
as well as in the frequency of observations above fixed mid to
high thresholds (5–15 m s−1, see Fig. 2). Trend results are robust
to changes in the station selection method and parameters (see
Supplementary Table S1).
If we assume that the primary cause of the surface wind stilling is
a slowdown in atmospheric general circulation and/or a weakening
in synoptic weather activity, then a stilling trend should also
show up (1) in the reanalyses of three-dimensional wind fields
of the National Centers for Environmental Prediction/National
Center of Atmospheric Research (NCEP/NCAR) and European
Center for Medium-range Weather Forecast (ECMWF)—because
observations of upper-air winds are used in the reanalysis procedure
and there is a high connectivity between winds at 10 m and
at 850 hPa, see Supplementary Fig. S2, (2) in upper-air wind
observations from rawinsonde and (3) in geostrophic winds
deduced from pressure gradients.
NCEP/NCAR reanalysis19 does not exhibit any trend in surface
(10 m) winds over land (Fig. 1b), as found in previous studies2,4,11.
The recent ECMWF ERA-interim reanalysis20 exhibits negative
trends with magnitudes between 10% (for North America) and
50% (for Europe) of the observed ones over the past two decades,
with quite different spatial patterns (Fig. 1c, Supplementary
Fig. S1 and Table S1). Over Australia similar weak trends
(−0.02 m s−1decade−1) were obtained by ERA-interim as by ERA40
in a previous study4. In contrast the inter-annual variability
of surface winds is fairly well reproduced in both reanalyses
(Supplementary Figs S2 and S3). This indicates that part of the
wind trend changes, captured in the ERA-interim reanalyses,
is due to large-scale circulation changes, but the reanalysis
NATURE GEOSCIENCE |ADVANCE ONLINE PUBLICATION |www.nature.com/naturegeoscience 1
© 2010 Macmillan Publishers Limited. All rights reserved.

LETTERS NATURE GEOSCIENCE DOI: 10.1038/NGEO979
Europe
Central Asia
East Asia
North America
a
c
b
¬5.00
¬1.00
¬0.50
¬0.20
¬0.10
¬0.05
0.05
0.10
0.20
0.50
1.00
5.00
¬1.00
¬0.50
¬0.20
¬0.10
¬0.05
0.05
0.10
0.20
0.50
1.00
5.00
¬5.00
¬5.00
¬1.00
¬0.50
¬0.20
¬0.10
¬0.05
0.05
0.10
0.20
0.50
1.00
5.00
Wind speed trend (m s
¬1
per decade)
Wind speed trend (m s¬1 per decade) Wind speed trend (m s¬1 per decade)
Figure 1 |Observed and reanalysis surface wind speed trends. a, 30-year surface (1979–2008) wind speed linear trend calculated over all of the available
observations and each of the selected stations, in m s−1perdecade. The specific regions studied in this Letter are shown by rectangles delimiting the areas
over which statistics are calculated. b, NCEP/NCAR reanalyses trends calculated using mean daily 10 m wind speed values available over a grid of 193×47
grid points over the Northern Hemisphere (data available at http://www.cdc.noaa.gov/data/gridded/data.ncep.reanalysis.html). c, The same as in bfor
ERA-interim reanalyses trends calculated over the past 20 years alone. The area boundaries for the four regions of focus are: Europe 20◦W–40◦E,
30◦–75 N, 276 stations; Central Asia 40◦–100◦E, 30◦–75◦N, 96 stations; Eastern Asia 100◦–160◦E, 30◦–75◦N, 190 stations; North America, 170◦–50◦W,
30◦–75◦N, 170 stations. The South Asia area quoted in the main text covers 40◦–160◦E, 0◦–30◦N, 40 stations.
models/procedures have deficiencies and/or missing key processes
that prevent the surface wind trends from being fully resolved2,4,11.
Land-use changes, not taken into account in reanalyses, are
potential candidates of such processes.
Upper-air (850 hPa and above) wind speed observations from
rawinsonde data do not either exhibit such a systematic wind
decline pattern (Fig. 3), and even show large regions with increasing
trends (Western Europe, North America). At nearly collocated sites,
winds normalized by their grand mean have trends with much
higher amplitudes at the surface than above (Fig. 3c). However,
regionally coherent negative trends are found over Eastern Asia
at 850 hPa, where surface trends could be linked with upper-air
trends, as a result of regional monsoon decline due to climate
change and air pollution3. However, reanalyses over China do not
exhibit trends, which remains unexplained and points towards
uncertainties in models.
2NATURE GEOSCIENCE |ADVANCE ONLINE PUBLICATION |www.nature.com/naturegeoscience
© 2010 Macmillan Publishers Limited. All rights reserved.

NATURE GEOSCIENCE DOI: 10.1038/NGEO979 LETTERS
1980 1990 2000
Year
0.1
1
10
100
1980 1990 2000
0.1
1
10
100
1980 1990 2000
0.1
1
10
100
Frequency (%) Frequency (%)
1980 1990 2000
Year
Year
Year
0.1
1
10
100
Frequency (%) Frequency (%)
u ≥ 1 m s¬1
u ≥ 3 m s¬1
u ≥ 5 m s¬1
u ≥ 7 m s¬1
u ≥ 9 m s¬1
u ≥ 11 m s¬1
u ≥ 13 m s¬1
u ≥ 15 m s¬1
a
c
b
d
Figure 2 |Wind speed distribution evolution. a–d, Evolution, as a function of year, of the annual frequency of wind exceeding a given threshold (various
curves, see legend), for each of the four regions identified in Fig. 1: Europe (a), Central Asia (b), Eastern Asia (c), North America (d). The numbers take into
account all stations in the domains together, as well as all hours and seasons. The linear regression trend coefficients, respectively for the frequency (in
% perdecade) of winds stronger than 1, 3, 5, 7, 9, 11, 13 and 15m s−1are: Europe: 0,−1,−5,−7,−11,−12,−11,−12; Central Asia:
0,−6,−13,−18,−23,−24,−22,−23; Eastern Asia: 2,−4,−10,−15,−19,−23,−30,−37; Northern America: −1,−2,−4,−5,−3,−3,−3,−11.
Geostrophic winds estimated from observed sea-level pressure
gradients also show a negligible stilling trend2,8 (Supplementary
Fig. S3). However, their interannual variations are consistent with
observed surface winds. These findings suggest that large-scale wind
changes should not be the dominant drivers of observed surface
wind decline.
Despite our stringent station selection procedure of rejecting
stations with large data gaps and obvious breaks (see the Methods
section and Supplementary Information), some time series may
still exhibit heterogeneities, which could explain a widespread
wind stilling trend only if they would be coherent over all
regions, which is unlikely. Systematical instrumental drifts, such
as anemometer performance degradation, especially for near-calm
winds, have been suggested17, but anemometers were replaced
over the past decades (O. Mestre and S. Jourdain, Météo-France,
personal communication), and trends are more pronounced for
stronger winds. A systematic change of observation systems
or round-off procedures18 also could have introduced artificial
trends. However, at most stations the decreasing wind trend was
gradual throughout the past 30 years, as shown for instance in
Fig. 4a for the Central Asian stations with wind trends in the
lowest tercile of the data. It is therefore unlikely that generalized
instrumental heterogeneities and observational procedures explain
the systematic trends.
These findings suggest that changes in surface processes
could play a major role in the surface wind stilling. Surface
winds are sensitive to changes in (1) surface roughness and
(2) sensible heat fluxes modifying vertical momentum fluxes
through boundary layer convection. Wind trends do not exhibit
strong diurnal variations (see Supplementary Table S1), suggesting
that roughness changes should dominate over sensible heat flux
changes as a cause of stilling.
Increase of surface roughness can be associated with factors
such as urbanization, growth of forests, changes in trees and forest
distribution or changes in agricultural practices. To estimate the
sensitivity of surface wind to roughness changes, we carried out
sensitivity simulations with the MM5 model21. Starting from a
control simulation (with standard roughness height z0c), changes
in roughness height z0(from a factor 1 to 2) have been applied in
the model surface boundary conditions during a single year (2007)
over an area covering Eastern Europe and Central Asia, where
the negative wind trends are most pronounced. When averaged
over the area considered (Supplementary Fig. S4), the 10 m wind
speed decrease δUbetween an enhanced roughness simulation and
a control simulation fits the relation δU= −aln(z0/z0c), where
a=0.48 m s−1when surface roughness is uniformly changed and
a=0.37 m s−1when only grassland roughness is changed and
the effect is averaged over grassland areas (as defined by MM5
land-cover data). Thus, about a doubling of roughness height would
lead to a decrease of 10 m wind by 0.26–0.33 m s−1, which is in the
range of observed wind decreases over the past three decades (see
Supplementary Table S1). Over Central Asia, where trends are most
pronounced, a tripling of the roughness height would be required
to fully reach observed stilling, which seems fairly unrealistic.
NATURE GEOSCIENCE |ADVANCE ONLINE PUBLICATION |www.nature.com/naturegeoscience 3
© 2010 Macmillan Publishers Limited. All rights reserved.

LETTERS NATURE GEOSCIENCE DOI: 10.1038/NGEO979
North America
Europe
Central Asia
East Asia
100
200
300
400
500
600
700
800
900
1,000
Pressure (hpa)
100
200
300
400
500
600
700
800
900
1,000
Pressure (hpa)
5.00
1.00
0.50
0.20
0.10
0.05
¬0.05
¬0.10
¬0.20
¬0.50
¬1.00
¬5.00
¬0.2 ¬0.1 0 0.1 0.2 0.3 0.4 0.5 ¬0.06
Europe
Central Asia
Eastern Asia
North America
¬0.04 -0.02 0 0.02
Normalized wind speed trend (per decade)
Wind speed trend (m s¬1 per decade)
b
a
c
Wind speed trend (m s¬1 per decade)
Figure 3 |Upper-air wind speed trends. a, Trends of the 850hPa wind measured by rawinsonde data, over stations having at least half of the yearly data
over more than 15 years during the two-decade period between 1979 and 2008, in m s−1perdecade. b, Mean vertical profile of wind speed trend obtained
from monthly averaged rawinsonde measurements (850 hPa and above) and closest surface site, averaged over all sites in each region of Fig. 1a. c, The
same as in bfor wind speed trends normalized by mean wind speed at each site. Surface values are arbitrarily set to 1,015 hPa pressure in the profile.
An average increase of about 1.1% peryear of forest carbon sink
at mid to high latitudes of the Northern Hemisphere band has
been deduced from two-decade-long remote-sensed normalized
difference vegetation index (NDVI) data22. Over Russia, estimates
of the yearly carbon sink to carbon pool ratio vary between 1.2
and 1.3%. Using a theory developed to link tree properties and
roughness23,24, applied to areas with only sparse roughness elements
(which is generally the case for anemometer neighbourhoods), the
roughness height should scale as z0∼hλ1.33, where his the height
and λis the frontal area index of roughness elements. Assuming
an isotropic development of the volume Vof roughness elements
and their constant density per area, hscales as V1/3and λscales as
V2/3, implying that z0scales as V1.22. Conversely, assuming only an
increase in the density of roughness elements without any change
in their volume, the increase of z0should scale as the increase of
V. From this scaling one can deduce that the increase of biomass
over Russia would have induced, over 30 years, an increase of
roughness height of 36–50%, explaining 25–40% of the observed
wind stilling. This is also supported by the cropland abandonment
in southwestern Russia25. The contribution of biomass growth to
wind stilling should be higher in other areas such as Europe or
China, where forest growth reaches 1.5% per year (refs 26,27). In
such cases, the induced wind decline could reach between 40% and
60% of the observed one over the past three decades.
Several studies based on the NDVI showed a widespread
increase of vegetation in many northern areas28. Using the Global
Inventory Modelling and Mapping Studies (GIMMS) database29,
obtained from the Advanced Very High Resolution Radiometer
(AVHRR) remote-sensed observations over 1982–2006, we found
that 62% of wind sites witnessed a positive spring–summer NDVI
trend (Fig. 4b). Moreover, the negative wind trend has a median
amplitude about three times larger over sites with a NDVI trend in
the upper 10% than over sites in the lower 10% of its distribution
(Fig. 4c). This also supports the influence of vegetation increase
on surface winds, but a quantitative relationship between NDVI
and roughness is difficult to establish. Note also that the declining
wind trends are found across the entire day (Supplementary Table
S1), suggesting that dynamic roughness changes dominate over
sensible heat flux changes.
Therefore, atmospheric circulation changes captured by the
ERA-interim reanalysis explain between 10 and 50% of wind
stilling in the northern mid-latitudes, and our analysis suggests
that 25–60% could be due to a roughness increase due to
vegetation. Urbanization may be an extra roughness increase
factor, although recent observations do not directly support
this mechanism8. However, even a gross estimate of roughness
increase due to urbanization could not be found. To formally
attribute the causes of declining winds, a sensitivity analysis
4NATURE GEOSCIENCE |ADVANCE ONLINE PUBLICATION |www.nature.com/naturegeoscience
© 2010 Macmillan Publishers Limited. All rights reserved.

NATURE GEOSCIENCE DOI: 10.1038/NGEO979 LETTERS
1
1980 1985 1990 1995 2000 2005
2
3
4
5
6
¬0.3
¬0.2
¬0.1
0
0.1
¬0.01 0 0.01 0.02
0.500
0.100
0.050
0.020
0.010
0.005
¬0.005
¬0.010
¬0.020
¬0.050
¬0.100
¬0.500
Spring¬summer NDVI trend (per decade)Year
Wind speed trend (m s¬1 per decade)
Annual mean wind speed (m s¬1)
ac
b
North America
Europe
Central Asia
East Asia
Spring¬summer NDVI trend (per decade)
80%
60%
40%
20%
Figure 4 |Surface wind trends and their relationship with NDVI trends. a, Time evolution of annual average surface wind speed for each of the stations
with the trend in the lower tercile of the wind trend distribution over the Central Asia area. Each curve corresponds to a station. The mean trend is
−0.31 m s−1decade−1.b, Spring–Summer (April–September) trends (in decade−1), of the average over a 24×24 km area around the surface wind stations
of the NDVI (as obtained from the GIMMS AVHRR product). c, Median surface wind trends over the stations of each decile of the NDVI trend distribution,
together with the corresponding wind trend distribution (20–40% and 60–80% in yellow, 40–60% in red), versus the median of each decile of NDVI
trends. The regression line for the median (y=−0.104–2.65x) is shown.
using a model of surface wind speeds driven by all potential
variables is required.
An important question is whether wind power energy produc-
tion is likely to be affected by wind stilling in the future. Over the
past 30 years surface winds underwent an average 10% decrease,
with larger trends at stronger winds (Fig. 2). As wind electricity
production is much more efficient at stronger winds, a continuation
of such trends would lead to a major loss of wind power production.
However, wind power is not taken at the surface but between
50 and 100 m, where trends should be weaker according to our
three-dimensional analysis. In any case, the strong influence of
land-cover change on wind speed trends is good news because
land use can much more easily be controlled locally than the
large-scale circulation.
Methods
Surface winds. Observation data sets come from the Research Data Archive, which
is maintained by the Computational and Information Systems Laboratory at NCAR
(sponsored by the National Science Foundation). The original data were taken
from the Research Data Archive (http://dss.ucar.edu).
Surface winds are obtained from two global data sets of this archive, gathering
hourly or three-hourly meteorological information on several parameters including
wind speed and direction from anemometers. The first data set (DS463.2) covers
the period 1901–2003. The second data set (DS461.0) covers the period 2000–2008.
Both contain hourly or three-hourly reports from about 10,000 observation sites
worldwide. The database includes data originating from various sources such as
synoptic, airways, Meteorological Routine Weather Report and Supplementary
Marine Reporting Station. A full description of the data sets can be found on
the websites http://dss.ucar.edu/datasets/ds463.2/docs/td9956.200301.pdf and
http://dss.ucar.edu/datasets/ds461.0/docs/. In particular, care was taken to correct
some heterogeneities induced by changes in wind speed units in this data set (see
related documentation on this latter website).
A stringent objective procedure is applied here to remove stations with gaps
or heterogeneities in the data, as described in the Supplementary Information,
leaving a subset of 822 worldwide stations, most of which are located in the
Northern Hemisphere.
Upper-air winds from rawinsondes. We used the rawinsonde monthly averages
of wind gathered in the Integrated Global Radiosonde Archive30 (IGRA). These
data were available for comparison with surface data over the three-decade period
from 1979 to 2008, and we used values taken at standard levels of 850 hPa, 700hPa,
500 hPa and 200hPa. A site selection procedure is also applied as for surface data,
see Supplementary Information. IGRA also includes surface wind speed taken near
the site. However, we did not use these data because the pairing would have led to
less complete profiles than if using the NCAR surface wind data and a criterion of
300 km for distance. We however checked that the IGRA surface wind speed had
similar trends to the NCAR data.
NCEP/NCAR and ERA-interim reanalyses. Reanalyses issued at
NCEP/NCAR are used, and made available at a 2.5×2.5◦resolution on the
NATURE GEOSCIENCE |ADVANCE ONLINE PUBLICATION |www.nature.com/naturegeoscience 5
© 2010 Macmillan Publishers Limited. All rights reserved.

LETTERS NATURE GEOSCIENCE DOI: 10.1038/NGEO979
http://www.cdc.noaa.gov/data/gridded/data.ncep.reanalysis.html website. Only
the 4 ×daily data set of wind at 10 m is used, for the Northern Hemisphere over
the period 1979–2008. The new ECMWF ERA-interim20 is a global atmospheric
reanalysis covering the period from 1989 until real time. The production
of ERA-interim began in summer 2006. Increased computer power enabled
several enhancements of ERA-interim over ERA-40. In particular, ERA-interim
uses a model version that was operational for weather forecast in the autumn
of 2006, with improved model physics. It also uses a 12-h 4D-VAR data
assimilation system, a horizontal resolution of ∼80 km, a more extensive use
and better treatment of satellite radiances, a better formulation of background
error constraints and a new humidity analysis. The long-term homogeneity
has improved substantially over that of ERA-40. The reader is referred to
http://www.ecmwf.int/research/era/do/get/index for a description of reanalysis
activities and products at ECMWF.
NDVI from AVHRR. NDVI data from the GIMMS database29, obtained from
the AVHRR remote-sensed observations, were used. These data cover the 25-year
period 1982–2006 and were carefully calibrated. They were originally provided
on a 8 km ×8 km grid, but averages over available NDVI data over the nine cells
surrounding the station were used for plotting Fig. 4b,c. We then calculated linear
trends from six-month NDVI averages (April–September).
Received 22 April 2010; accepted 9 September 2010;
published online 17 October 2010
References
1. Klink, K. Trends in monthly maximum and minimum surface wind speeds in
the coterminous United States, 1961 to 1990. Clim. Res. 13, 193–205 (1999).
2. Smits, A., Klein-Tank, A. M. G. & Können, G. P. Trends in storminess over the
Netherlands, 1962–2002. Int. J. Climatol. 25, 1331–1344 (2005).
3. Xu, M. et al. Steady decline of East Asian monsoon winds, 1961–2000:
Evidence from direct measurements of wind speed. J. Geophys. Res. 111
doi:10.1029/2006JD007337 (2006).
4. McVicar, T. R. et al. Wind speed climatology and trends for
Australia, 1975–2006: Capturing the stilling phenomenon and
comparison with near-surface reanalysis output. Geophys. Res. Lett. 35
doi:10.1029/2008GL035627 (2008).
5. Pryor, S. C., Barthelemie, R. J. & Schoof, J. T. Interannual variability of wind
indices across Europe. Wind Energy 9, 27–38 (2005).
6. Roderick, M. L., Rotstayn, L. D., Farquhar, G. D. & Hobbins, M. T. On the
attribution of changing pan evaporation. Geophys. Res. Lett. 34, L17403 (2007).
7. Roderick, M. L., Hobbins, M. T. & Farquhar, G. D. Pan evaporation trends
and the terrestrial water balance II. Energy balance and interpretation.
Geogr. Compass 3, 761–780 (2009).
8. Rayner, D. P. Wind run changes: The dominant factor affecting pan
evaporation trends in Australia. J. Clim. 20, 3379–3394 (2007).
9. Hua, G., Ming, X. & Qi, H. Changes in near-surface wind speed in China:
1969–2005. Int. J. Climatol. 30 doi:10.1002/joc.2091 (2010).
10. Brazdil, R. et al. Climate fluctuations in the Czech Republic during the period
1961–2005. Int. J. Climatol. 29, 223–242 (2009).
11. Pryor, S. C. et al. Wind speed trends over the contiguous United States.
J. Geophys. Res. 114, D14105 (2009).
12. Pirazzoli, P. A. & Tomasin, A. Recent near-surface wind changes in the Central
Mediterranean and Adriatic areas. Int. J. Climatol. 23, 963–973 (2003).
13. Aristidi, E. et al. An analysis of temperatures and wind speeds above Dome C,
Antarctica. Astron. Astrophys. 430, 739–746 (2005).
14. Lynch, A. H., Curry, J. A., Brunner, R. D. & Maslanik, J. A. Toward an
integrated assessment of the impacts of extreme wind events on Barrow, Alaska.
Bull. Am. Meteorol. Soc. 85, 209–221 (2004).
15. Lu, J., Vecchi, G. A. & Reichler, T. Expansion of the Hadley cell under global
warming. Geophys. Res. Lett. 34, L06805 (2007).
16. Seidel, D. J., Fu, Q., Randel, W. J. & Reichler, T. J. Widening of the tropical belt
in a changing climate. Nature Geosci. 1, 21–24 (2008).
17. DeGaetano, A. T. Identification and implications of biases in US surface wind
observation, archival, and summarization methods. Theor. Appl. Climatol. 60,
151–162 (1998).
18. McKee, T. B., Doesken, N. J., Davey, C. A. & Pielke, R. A. Climate data
continuity with ASOS: Report for period April 1996 through June 2000, 82,
Colo. Clim. Cent., Fort Collins (2000).
19. Kalnay, E. et al. The NCEP/NCAR 40-year reanalysis project. Bull. Am.
Meteorol. Soc. 77, 437–471 (1996).
20. Uppala, S., Dee, D., Kobayashi, S., Berrisford, P. &
Simmons, A. Towards a climate data assimilation system:
Status update of ERA-Interim. ECMWF Newslett. 115,
12–18 (2008) (Available at: http://www.ecmwf.int/publications/newsletters/).
21. Dudhia, J. A nonhydrostatic version of the Penn State/NCAR mesoscale
model: Validation tests and simulation of an Atlantic cyclone and cold front.
Mon. Weath. Rev. 121, 1493–1513 (1993).
22. Myneni, R. B. et al. A large carbon sink in the woody biomass of Northern
forests. Proc. Natl Acad. Sci. 98, 14784–14789 (2001).
23. Raupach, M. R. Drag and drag partition on rough surfaces. Bound. Layer
Meteorol. 60, 375–395 (1992).
24. Raupach, M. R. Simplified expressions for vegetation roughness length and
zero-plane displacement as a function of canopy height and area index.
Bound. Layer Meteorol. 71, 211–216 (1994).
25. Vuichard, N. et al. Carbon sequestration due to the abandonment of agriculture
in the former USSR since 1990. Glob. Biogeochem. Cycles 22, GB4018 (2008).
26. Ciais, P. et al. Carbon accumulation in European forests. Nature Geosci. 1,
425–429 (2008).
27. Kauppi, P. E. et al. Returning forests analysed with the forest identity.
Proc. Natl Acad. Sci. 103, 17574–17579 (2006).
28. Donohue, R. J., McVicar, T. R. & Roderick, M. L. Climate-related changes
in Australian vegetation cover as inferred from satellite observations for
1981–2006. Glob. Change Biol. 15, 1025–1039 (2009).
29. Tucker, C. J. et al. An extended AVHRR 8-km NDVI data set compatible
with MODIS and SPOT vegetation NDVI data. Int. J. Remote Sens. 26,
4485–5598 (2005).
30. Durre, I., Russell, S., Vose, & Wuertz, D. B. Overview of the integrated global
radiosonde archive. J. Clim. 19, 53–68 (2006).
Acknowledgements
We thank X. Wang for his help with extraction of NDVI fields at the stations. We
benefited from fruitful discussions with F-M. Bréon at LSCE.
Author contributions
R.V. designed the experiments and carried out the observation analysis, J.C. did
all model experiments and analysis of outputs, as well as NCEP/NCAR reanalysis
calculations. J-N.T. extracted and analysed all data from ERA-interim reanalyses. P.Y.
initially suggested systematic wind trends in observations, and P.C. suggested that
land-cover changes due to vegetation could be a major driver of wind stilling and
helped design the experiments with NDVI. All co-authors substantially contributed to
the paper writing.
Additional information
The authors declare no competing financial interests. Supplementary information
accompanies this paper on www.nature.com/naturegeoscience. Reprints and permissions
information is available online at http://npg.nature.com/reprintsandpermissions.
Correspondence and requests for materials should be addressed to R.V.
6NATURE GEOSCIENCE |ADVANCE ONLINE PUBLICATION |www.nature.com/naturegeoscience
© 2010 Macmillan Publishers Limited. All rights reserved.