Over the past 30 years, observations indicate a decline of about −0.3 m/s in the northern mid-latitudes land surface wind speed. The picture is less conclusive for the Southern Hemisphere and over the oceans. Such a stilling can affect surface evaporation and climate feedback processes, and may impact technical applications such as wind power. Using an atmospheric global climate model, we perform sensitivity experiments for the period 1870–2005 to assess the role of changing roughness length, aerosol emissions, sea surface temperature, and greenhouse gas concentrations in surface wind speed changes. The wind speed trends simulated by the model generally underestimate the observed trends (land and ocean). Over land, the model can reproduce the observed stilling by increasing the roughness length by a factor of 1.2 to 4.9, depending on region. The other forcings examined can also decrease the 10 m wind speeds (up to 15% of observed values in Europe), particularly those related to increasing aerosol emissions (up to −0.2 m/s in India). Compared to observations, the simulated impact of climate forcings on global wind speeds over land and ocean is however small and not always significant.
 According to Vautard et al. , 25 to 60% of the northern mid-latitude stilling in the past 30 years may be attributed to a recent increase in vegetation cover [e.g.,Kauppi et al., 2006; Ciais et al., 2008] and associated changes in roughness length, whereas up to 50% may be attributed to atmospheric circulation changes. The recent increase in vegetation cover observed in the Northern Hemisphere is mostly due to improved silvicultural practices and enhanced fertility [e.g., Kauppi et al., 2006; Ciais et al., 2008]. However, given the lack of objective information on past changes in the Northern Hemisphere's roughness length, Vautard et al. base their conclusions on regional sensitivity studies for Eurasia (20°–80°E, 30–70°N). In addition, only few studies investigate the contribution from atmospheric circulation changes: Most of them focus on China and stress the dominant role of sea-surface temperatures (SSTs) in controlling the surface wind speed, in particular the land-sea temperature gradient, itself sensitive to aerosol and greenhouse gas concentrations [e.g.,Xu et al., 2006; Li et al., 2010; Guo et al., 2011]. For the US, Klink  also suggests that a reduced meridional temperature gradient in response to global warming could partly explain the observed wind stilling. Other possible causes for the wind stilling have also been mentioned [e.g., McVicar et al., 2012, Section 2.4], such as the recent increase in urbanization [Xu et al., 2006], but no dataset covering the roughness increase due to urbanization is currently available to further investigate the issue. In addition, the increasing trend in available soil water (e.g., driven by changes in precipitation and irrigation), which potentially increases the latent heat flux and decreases the sensible heat flux, has also been mentioned [Shuttleworth et al., 2009], but is not discussed in this study.
 In our paper, we extend previous sensitivity studies to the global scale, investigating the response of the 10 m wind speed to friction changes (roughness length from vegetation) on the one hand, and to changes in atmospheric forcings (sea surface temperatures, aerosol emissions, and greenhouse gas concentrations) on the other hand.
 We perform climate simulations using the fifth generation of the atmospheric global climate model (GCM) ECHAM5 [Roeckner et al., 2003], at horizontal resolution T42. The methodology follows Bichet et al.  and Folini and Wild . We use a version of ECHAM5 that is coupled to the fully interactive Hamburg Aerosol Module (HAM) [Stier et al., 2005], which predicts the evolution of seven interacting, internally and externally mixed lognormal aerosol modes. A double-moment cloud microphysics scheme is used [Lohmann et al., 2007] that couples to the size-resolved aerosol scheme of HAM and predicts the mass mixing ratios and number concentrations of cloud droplets and ice crystals.
 We conduct a series of experiments driven by monthly mean observed SSTs and sea-ice concentrations [Rayner et al., 2003], accounting for different atmospheric forcings. Other forcings used include time varying monthly means of the total solar irradiance (TSI) [Solanki and Krivova, 2003] and of stratospheric optical depth due to aerosols from explosive volcanoes [Sato et al., 1993], as well as time varying annual mean greenhouse gas concentrations taken from observations until 2000 and from the IPCC A1B scenario for 2001–2005 (carbon dioxide, methane, nitrous oxide, ozone, and chlorofluorocarbons). Aerosol emissions of sulfur dioxide, black carbon, and particulate organic matter are taken from the Japanese National Institute for Environmental Studies (NIES) [Roeckner et al., 2006; Stier et al., 2006; Nozawa et al., 2007]. They include geographically resolved, time varying monthly mean emissions from wildfires, agricultural burning, and domestic fuel-wood consumption, as well as time varying annual mean emissions from fossil fuel consumption. Stratospheric ozone and land use are both kept at climatological values.
 We perform thirty transient experiments running from 1870 to 2005. Thirteen experiments correspond to the control runs (CTRL, “all forcings runs”), for which all the forcings are time varying except for the roughness length which is held constant. Ten experiments are identical to CTRL except that aerosol emissions (anthropogenic and natural, including explosive volcanoes) are held constant at their 1870 values (referred to as AEC), four are identical to CTRL except that SSTs are held constant at their climatological values averaged over the time period 1870–1900 (referred to as SSTC), and three are identical to CTRL except that both, SSTs and aerosol emissions are held constant (referred to as AESSTC). To suppress the “noise” from individual simulations and assess their natural variability, we calculate ensemble means and quantify their uncertainties via the computation of standard deviations. We also compute the 10 m wind speed trends and assess their statistical significance.
 The sensitivity to increasing roughness length is assessed by conducting 40 years long simulations (using climatological SSTs and climatological aerosol emissions for the period 1860–1900) with the same model configuration, where the monthly climatological values of the land surface roughness (vegetation only) prescribed in the model are multiplied by a factor of 1.5, 2 and 4, respectively. We analyze the last 30 years mean. These simulations are a first step to understand the impact of increasing roughness length, even though the changes in roughness length in the “real world” are both geographically and time dependent. The simple approach is justified as the current knowledge on historic roughness length changes appears too vague to justify transient simulations.
3.1. Roughness Length
 The response of the 10 m wind speed to a uniform doubling of the vegetation roughness length over land shows strong regional differences, as can be taken from Figure 1. As listed in Table 1(columns 4–6), globally increasing the vegetation roughness length in ECHAM5-HAM by a factor of 1.5, 2 and 4 decreases the global land annual 10 m wind speed on average by −0.15, −0.26 and −0.55 m/s, respectively. Within this range, the relation is approximately linear at global scale as well as in the selected regions. Assuming this linear approximation and based on the results obtained with roughness length multiplied by 2 (Table 1, column 5), we find that in order to reproduce the observed wind stilling (Table 1, column 3) by modifying the roughness length only, this latter should increase by a factor of 1.2 to 4.9 over the past 30 years, depending on the region under consideration (Table 1, column 7).
Table 1. Wind Speed Trends and Anomalies (m/s in 30 Years)a
Wind Speed Trends (m/s in 30 Years)
Wind Speed Anomalies (m/s)
Factor Needed to Increase Z0 in ECHAM5-HAM to Match Observations
Wind Speed Trend Differences (m/s in 30 Years)
(1.5 × Z0 − Z0)
(2 × Z0 − Z0)
(4 × Z0 − Z0)
Columns 2 and 3 compare the annual trends in 10 m wind speed as simulated in the CTRL ensemble mean (30 years trend over the time period 1975–2005) and as observed (30 years trend over or scaled to the time period 1975–2005). Columns 4–6 show wind speed anomalies (averaged over 30 years) between the simulations with increased roughness length (Z0 × 1.5, Z0 × 2 and Z0 × 4) and the simulations with climatological roughness length (Z0). Column 7 estimates the factor by which roughness length needs to increase in order to reach the observed stilling, assuming a linear approximation and based on the results obtained with roughness length multiplied by 2 (column 5). Columns 8–9 show the differences between wind speed trends (1975–2005) simulated in the CTRL and the AEC and SSTC ensemble means, respectively. The values indicated in “bold” are statistically significant at 80%. The area boundaries for the regions of focus are described in Figure 1. Observations are from (1), Vautard et al. ; (2), Guo et al. ; (3) Roderick et al. ; (4), Wentz et al. ; (5) McVicar et al. ; (6), Donohue et al. ; (7), Troccoli et al. ; (8), McVicar et al. ; and (9), Oguntunde et al. .
 These results confirm the results by Vautard et al.  as obtained with a regional climate model and extend their findings to further regions and the global scale.
 Consistent with boundary layer theory, most of the impacts of roughness length changes will be local, i.e., occur in the same region as the roughness length changes. Regional changes in vegetation should thus imply a corresponding pattern of wind-speed changes, modulated by the regionally varying response (Figure 1). While observations indicate a recent vegetation increase (and associated increase in roughness length) in most of the Northern Hemisphere [e.g., Kauppi et al., 2006; Ciais et al., 2008], it appears unlikely that roughness length changes alone are able to explain the observed wind stilling in all the regions (Table 1, column 7). Other forcings may therefore also play a role [see McVicar et al., 2012, Section 2.4], which could include changes in SSTs, aerosol emissions, and greenhouse gas concentrations.
3.2. SSTs, Aerosol Emissions, and Greenhouse Gas Concentrations
 SSTs, aerosol emissions, and greenhouse gas concentrations can affect the atmospheric circulation and stability, and thereby surface wind speeds [e.g., Ramanathan et al., 2001; Meehl et al., 2007]. In our modeling framework, these factors are assessed using simulations with prescribed SSTs. Therefore, these sensitivity experiments evaluate the atmosphere-only impact of greenhouse gases and aerosols on wind speed. Note also that in this set-up, the effects of previous greenhouse gas and aerosol emissions may partly be encapsulated in the transient SST evolution.
 Since 1870, the CTRL ensemble mean (using all time-varying forcings but climatological roughness length) exhibits considerable decadal variations in annual-mean 10 m wind speed (Figure 2, black curves). In addition, it also exhibits long-term wind trends after the 1950s, which are either negative (e.g., Southern Hemisphere land and China) or positive (e.g., India). Over the past 30 years, these trends are generally smaller than indicated by observations (Table 1, column 2–3), but statistically significant at 80% in Central Eurasia, China and Sub-Sahara (Table 1, column 2, “bold” entries, and Figure S1 in the auxiliary material).1 Note that these trends are roughly 5 to 15 times smaller in all regions when compared to observations, so that the effect of SST variability, as well as aerosols and greenhouse gases may only explain a minor part of the observed wind speed trends at global scale.
 Based on the sensitivity experiments with varying (CTRL, black curves) and constant (SSTC, blue curves) SSTs, Figure 2shows that transient SSTs explain a considerable fraction of the decadal variations since 1870, but not the long-term trends. In the past 30 years, the difference between their respective linear trends (Table 1, column 9) shows that transient SSTs, as compared to constant SSTs, either increase (e.g., 0.08 m/s in the past 30 years in India) or decrease (e.g., −0.02 m/s in the past 30 years in Europe) the annual land 10 m wind speed, depending on the region. These trend anomalies, however, are significant at 80% only in small areas located mostly in Eurasia and Sub-Sahara (Table 1, column 9, “bold” entries, and Figure S2b in the auxiliary material).
 Based on the sensitivity experiments with varying (CTRL, black curves) and constant (AEC, red curves) aerosol emissions, Figure 2shows that in most regions, the atmosphere-only response of the annual 10 m wind speed to variable aerosol emissions is a decrease of up to −0.15, −0.1 and −0.08 m/s in India, Sub-Sahara and China, respectively, after about 1950. Over the past 30 years, the respective linear trends (Table 1, column 8) show that in most regions, the atmosphere-only response of the annual 10 m wind speed to aerosol emissions is a wind speed reduction of −0.13, −0.05 and −0.03 m/s in India, Sub-Sahara and China, respectively (average linear trends,Table 1, column 8). These trend anomalies are significant at 80% only in small areas located in Eurasia and Sub-Sahara (Table 1, column 8, “bold” entries, and Figure S2a in the auxiliary material). In most regions, aerosols have a larger impact in summer, decreasing the surface wind speeds by a maximum of −0.3 and −0.25 m/s in India since 1950 and 1975, respectively (not shown).
 Based on the sensitivity experiments with constant SSTs and constant aerosol emissions (AESSTC, green curves), Figure 2shows that after about 1950, the atmosphere-only response to the remaining external forcings (i.e., increasing greenhouse gas concentrations and TSI variations) is an increase in the 10 m wind speed. However, their impacts are relatively small after 1950 and almost negligible after 1975 (maximum impact of +0.03 m/s after 1975), suggesting that greenhouse gases combined with TSI have only a small atmosphere-only impact on the 10 m wind speed, and act mostly through the SSTs feedbacks. This result was expected since there is no oceanic feedback in these simulations.
 Our study shows that over land, increasing the roughness length clearly reduces the annual 10 m wind speed over most parts of the globe, with amplitudes that depend on the region (Table 1, columns 4–6, and Figure 1). However, the changes in roughness length required to reproduce the observations of the past 30 years are not necessarily realistic everywhere (Table 1, column 7), and other factors may have to be taken into account. The existence of such factors is illustrated by the black curves in Figure 2, showing that the 10 m wind speed simulated in the CTRL ensemble mean decreases after about 1950 in most of the northern hemispheric lands, despite being forced by climatological roughness length. According to Table 1 (column 2–3), the wind stilling trends simulated (CTRL) over the past 30 years represents 12% (ratio of the CTRL ensemble mean (Table 1, column 2) over observations (Table 1, column 3)) of the wind speed trend observed over the global oceans, and 3% of the wind speed trend observed over the northern hemispheric lands. However, regionally, this ratio reaches substantially higher values such as 20% in the tropical ocean, 15% in Europe, and 11% in China and Eastern Asia. Note that in China, this ratio reaches up to 20% in the summer averages [Guo et al., 2011].
 In our simulations, we find that over land, the factors decreasing the 10 m wind speed depend upon the region (Figure 2) and season (not shown) under consideration. For instance, whereas past SST variations are a dominant factor (aside from potential roughness length changes) in North America and Australia (compare black and blue curves in Figure 2), they have almost no impact in China. Similarly, aerosol emissions reduce the land 10 m wind speed over most of the globe, but with a magnitude that depends on the region (compare black and red curves in Figure 2). Some regions are also strongly affected by both forcings, such as in India, where SSTs increase the annual 10 m wind speed in the past few decades but aerosols “slow down” this increase (Figure 2). Note that over the ocean, the increasing 10 m wind speed trend (global and tropical) is also “slowed down” by the aerosol emissions (Figure 2).
 This regionally diverse response is consistent with the fact that SST-induced circulation changes have a regional character and can decrease or increase the 10 m wind speeds. In contrast, higher aerosol concentrations appear to generally reduce the 10 m wind speeds (land and ocean). This rather uniform response could be related to the role of atmospheric aerosols upon the stratification of the atmosphere. Whereas increasing aerosol emissions cool the surface, especially carbonaceous aerosols also warm the aerosol layer in the troposphere. This will increase the atmospheric density gradient between the surface and the troposphere, and thus slow down the atmospheric circulation [e.g.,Ramanathan et al., 2005]. In essence, increases in stratification have a tendency to decouple the planetary boundary layer from the atmospheric layers aloft, thereby reducing the influence of the upper-level winds at the surface. Our results show that the atmospheric stability simulated in the CTRL ensemble mean has increased since 1950 mostly in eastern Asia, Africa, Middle East, south-eastern North America, and South America. In eastern Asia and Africa, this is largely due to aerosols (Figure S3 in theauxiliary material). Simulated trends at different altitudes also suggest that aerosols decrease the global land and oceanic wind speed trends up to 850 hPa, which supports this hypothesis (Figure S4 in the auxiliary material).
 In addition, our results illustrate the land/ocean contrast, showing an increasing (+0.03 m/s in 30 years) and decreasing (−0.01 m/s in 30 years) trend in the global oceanic and land 10 m wind speed, respectively. According to our results, the increase in oceanic 10 m wind speed is mostly due to the impact of SST variations in the tropics and the Southern Hemisphere (see Figures S1 and S2b in the auxiliary material). As previously noted, this increasing trend over the global ocean would be more pronounced without the aerosol emissions.
 Finally, the latitudinal dependence pointed out by McVicar et al. , showing decreasing and increasing trends in the mid and high latitudes (∼>70°), respectively, is reproduced in our experiments in the Northern Hemisphere, although the results are not statistically significant (Figure S1 in the auxiliary material).
 In line with Vautard et al. , we find that over land, the recent increases in roughness length, for instance caused by increasing vegetation [e.g., Kauppi et al., 2006; Ciais et al., 2008], could well explain a significant fraction of the decrease in the terrestrial 10 m wind speed observed globally. This stresses the need for a better analysis of roughness length changes over the past decades, and motivates scenarios of such changes in the future. In addition to roughness length changes, we find that changes in other climate forcings contribute to the land wind stilling after 1950, by a magnitude representing up to 15% of the observed trends. In particular, we find increasing aerosol emissions to generally reduce the land 10 m wind speed, especially in summer. The simulated aerosol impacts are larger in Asia where the trends of aerosol emissions are particularly large, and where the simulated decrease in summer wind speeds amounts to up to 0.3 m/s after 1950. Over the oceans, we find that aerosol emissions “slow down” the simulated increasing trend of the 10 m wind speed. Nevertheless, the simulated response to variables SSTs and aerosol emissions remain a factor of 5 to 15 smaller than observed wind speed trends.
 The authors thank the MPI Hamburg for providing access to the ECHAM5-HAM code, colleagues at ETH Zürich and the Center for Climate System Modeling C2SM who contributed to model development, in particular Ulrike Lohmann, Sylvaine Ferrachat, and Grazia Frontosa, as well as Tanja Stanelle for extensive discussions on roughness lengths. This research has partly been supported by the NCCR Climate funded by the Swiss National Science Foundation. The simulations were done on the CRAY at the Swiss National Supercomputing Centre (CSCS) in Manno, Switzerland.
 The Editor thanks the two anonymous reviewers for assisting with the evaluation of this paper.