Previous studies indicated that surface wind speed over China declined during past decades, and several explanations exist in the literature. This study presents long-term (1960–2009) changes of both surface and upper-air wind speeds over China and addresses observed evidence to interpret these changes. It is found that surface wind over China underwent a three-phase change over the past 50 yr: (i) it step changed to a strong wind level at the end of the 1960s, (ii) it declined until the beginning of the 2000s, and (iii) it seemed to be steady and even recovering during the very recent years. The variability of surface wind speed is greater at higher elevations and less at lower elevations. In particular, surface wind speed over the elevated Tibetan Plateau has changed more significantly. Changes in upper-air wind speed observed from rawinsonde are similar to surface wind changes. The NCEP–NCAR reanalysis indicates that wind speed changes correspond to changes in geopotential height gradient at 500 hPa. The latter are further correlated with the changes of latitudinal surface temperature gradient, with a correlation coefficient of 0.88 for the past 50 yr over China. This strongly suggests that the spatial gradient of surface global warming or cooling may significantly change surface wind speed at a regional scale through atmospheric thermal adaption. The recovery of wind speed since the beginning of the 2000s over the Tibetan Plateau might be a precursor of the reversal of wind speed trends over China, as wind over high elevations can respond more rapidly to the warming gradient and atmospheric circulation adjustment.
Widespread surface wind speed declines have been observed from ground measurements over the past few decades in many tropical and midlatitude regions including China (McVicar et al. 2012, and references therein). These declines of surface wind speed may contribute to declines of atmospheric potential evaporation, as measured by pan evaporation (Chen et al. 2006; McVicar et al. 2012; Roderick et al. 2007), and the weakening trend in atmospheric sensible heat over some regions (Duan and Wu 2008; Yang et al. 2011a).
Recently, there has been great interest in the causes of the surface wind decline, as it is a key component of global climate change. Nevertheless, a widely accepted explanation has not been reached yet. By investigating normalized difference vegetation index (NDVI) data, Vautard et al. (2010) partly attributed the lowering of Northern Hemisphere atmospheric wind to an increase of surface roughness, but their study did not consider some specific cases such as one happening over the elevated Tibetan Plateau (TP) where wind speed considerably declined over the last 30 yr (Yang et al. 2011b) but the surface roughness change is not so significant. In Australia, McVicar et al. (2012) found that NDVI is slightly positively correlated with surface wind speed trends. The regional cooling in south-central China caused by air pollution was also speculated to be the cause of the weakening of the East Asia summer monsoon (Xu et al. 2006), which needs a further discussion. Alternatively, the weakening of atmospheric circulation under the background of global warming was considered to be the main factor contributing to the decline of surface wind in several studies (e.g., Duan and Wu 2009; Guo et al. 2011; Jiang et al. 2010; You et al. 2010; Zhang et al. 2009b).
These previous studies focused more on the wind speed changes over the globe and China during the past decades without including the most recent changes or addressing their elevation dependence. Calculating the trends of wind speed with respect to elevation in two mountainous regions in China and Switzerland, McVicar et al. (2010) uncovered the phenomenon that surface winds were declining more rapidly at higher elevations than lower elevations. Yang et al. (2011b) also pointed out that wind speed over the Tibetan Plateau during the last 30 yr declined much more than the one averaged over China, and therefore there is a need to give a detailed investigation on the elevation dependence of wind speed change and its cause. If so, this elevation dependence may imply the relationship with upper-air wind speed, which has not been taken into accounting in previous studies. Figure 3a of Vautard et al. (2010) showed decreasing upper-air (850 hPa) wind speed trends over the past three decades (1979–2008) in Asian regions, which is in contrast with the increasing trends over European–American regions. This difference was not discussed as Vautard et al. (2010) focused primarily on the European–American regions. Zhang et al. (2009a) also reported the decreasing upper-air wind speed trends during 1980–2006 over China, especially in the lower troposphere. However, a longer-term upper-air wind speed analysis would be helpful to further explore whether there is a relationship between the surface wind change and the upper-air wind change.
In this study, we aim to present a comprehensive analysis on the trends of wind speed over China during the past 50 yr. To achieve this, it is analyzed from following several aspects: (i) the observed trends of annual and seasonal mean surface wind speed; (ii) the spatial distribution of observed trends in surface wind speed; (iii) the elevation dependence of observed trends in surface wind speed; (iv) the trends of observed upper-air wind speed; and (v) the trends in latitudinal gradients of 500-hPa geopotential height and surface temperature. The results are given in section 4. The detailed description of the datasets and trend-calculating methods applied in this study is presented in section 2. As the quality and homogeneity of data are critical for long-term change analysis, we first present cross-validations to confirm the reliability of the derived wind speed trends in section 3, in terms of (i) comparison between Integrated Global Radiosonde Archive (IGRA) 850-hPa and China Meteorological Administration (CMA) surface wind speed and (ii) comparison between IGRA 850-hPa and CMA surface wind speed. In section 5 of concern and discussed are the causes of the wind speed changes over China. Conclusions are given in section 6.
2. Datasets and methods
a. CMA surface routine data
CMA provides for this study a long-term (from January 1960 to December 2009) routine dataset of surface wind speed observed at 10 m above the surface. At each CMA station, surface wind speed was measured 6-hourly [at 0200, 0800, 1400, and 2000 Beijing standard time (BST); BST = UTC + 8 h]. Figure 1 shows the distribution of CMA stations and their elevations. The elevations of the CMA stations vary from sea level up to 4800 m above mean sea level (MSL). There are a total of 125 stations with elevations greater than 2000 m MSL located in the TP area. This region exerts profound influences not only on the local climate and environment but also on the global atmospheric circulation through its thermal and mechanical forcing (Yeh and Gao 1979). Meanwhile, its environment is sensitive to global climate change (Liu and Chen 2000). Because of its unique environmental characteristics and close link to the upper-air atmospheric circulation, the wind speed changes over this region were highlighted and discussed herein.
b. IGRA upper-air radiosonde data
The radiosonde data gathered in IGRA are used to analyze the changes of upper-air wind in this study. The IGRA archive is integrated from different sources with a sequence of specialized quality assurance algorithms (Durre et al. 2006). The filled square symbols in Fig. 1 denote the IGRA (totaling 149) stations in China. These data are available for 1964–2009 for some stations in China; nevertheless, data over 1973–90 for many stations are missing. Hence we focused on 1990 onward. Monthly averaged values of upper-air wind speed at mandatory levels of 850, 700, 500, 300, and 200 hPa were used here.
c. Other datasets
To further explore the causations of the derived wind speed changes, several datasets are used in the study. First, the National Centers for Environmental Prediction–National Center of Atmospheric Research (NCEP–NCAR) reanalysis (Kalnay et al. 1996) was used. McVicar et al. (2008) showed that of the three common reanalysis outputs [NCEP–NCAR, NCEP/U.S. Department of Energy (DOE), and 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40)] tested, the NCEP–NCAR reanalysis was best able to capture observed near-surface wind speed trends during 1979–2001 for Australia. We therefore used NCEP–NCAR 500-hPa geopotential height data to examine whether the changes exist in the large-scale atmospheric circulation. Second, the Hadley Centre and Climate Research Unit (CRU) gridded near-surface temperature dataset, version 3 (HadCRUT3; Brohan et al. 2006; Jones et al. 1999; available at http://www.cru.uea.ac.uk/cru/data/temperature/), was used to further explore the changes in latitudinal gradient of surface warming rate. Third and finally, three climate indices were used: (i) Arctic Oscillation (AO); (ii) North Atlantic Oscillation (NAO); and (iii) Pacific decadal oscillation (PDO). AO is an annular mode of atmospheric circulation (Thompson and Wallace 1998). NAO is the difference of atmospheric pressure at sea level between the Icelandic low and the Azores high (Wallace and Gutzler 1981). AO and NAO are the dominant patterns that affect Northern Hemisphere climate variability, which are the most prevalent in winter at the mid–high latitudes. Both indices are available online (at http://www.cpc.ncep.noaa.gov/). PDO is the leading principal component of monthly sea surface temperature (SST) anomalies in the North Pacific Ocean (poleward of 20°N) (Mantua et al. 1997); it is available online (at http://jisao.washington.edu/pdo/PDO.latest). The period 1960–2009 was used in this study to match with the temporal extant of the CMA surface dataset. These indices are examined in terms of their relationship with the changes in surface wind speed.
d. Analysis procedures
To account for the irregularly distributed CMA stations, the area-weighted grid average is preferred in climate analysis. Duan (2009), however, showed that the station-averaged surface wind speed exhibits changes similar to those that are grid averaged, although the two averages have a systematic difference. We therefore used the simple station averaging to derive the regional mean wind speed changes. Based on the CMA surface wind speed dataset, we calculated the seasonal and annual mean values of wind speed at the four measuring times of day and their daily means at two spatial scales (China and TP). The station averages are calculated for the stations that have “good data records” (defined as data records that have more than 275 observations each year). There are 472 such CMA stations in China and 64 such stations in TP. Then, the linear trends of their time series are calculated by fitting a linear regression (ordinary least squares). The trend significance is tested using Student’s t test at the probability level of P < 0.05. At each station, the linear trend of surface or upper-air wind speed is also obtained based on the monthly mean deseasonalized values. The deseasonalization was performed through the process of removing the seasonal climatology from the time series.
In view of unrealistic surface wind speed trends calculated at a few stations, we fitted the dependence of wind speed trend on elevation using a robust linear regression to explore the impact of elevation on surface wind speed trend. The regression method (available in MATLAB Statistics Toolbox) employs an iteratively reweighted least squares algorithm with the weight at each iteration determined by applying a bisquare function to the residuals from the previous iteration (Huber 1981; Street et al. 1988). It can therefore minimize the effect of outliers and/or nonstationary variability compared to the ordinary linear least squares regression.
The wind speed changes are investigated through the pressure gradient force in the upper air. We defined a high-latitude band (HL; 45°–50°N) and a low-latitude band (LL; 20°–25°N). The upper-air wind speed over China, especially the TP region located between latitudes of 25° and 40°N, is largely determined by the pressure difference (or geopotential height gradient) between the two bands according to the geostrophic approximation theory. Using NCEP–NCAR reanalysis data, we calculated the geopotential height difference between the two bands (with longitudinal range of 70°–140°E) at the 500-hPa level as used to denote the geopotential height gradient. Correspondingly, we also examined the difference of surface temperature between the two latitude bands based on the CRU data.
3. Cross-validations of the derived wind speed changes
The quality and homogeneity of data are critical for long-term change analysis. Instrumental changes, observational drift, or other human-induced changes may yield unreal changes in time series. As far as we know, only air temperature is homogenized for the CMA stations, while surface wind speed is more sensitive to local environment so that its homogenization is more difficult and not done for the CMA stations. The rawinsonde wind data in IGRA have not yet been adjusted for inhomogeneity as well, and therefore one needs to be cautious when using these data for climate analysis (Durre et al. 2006). For instance, Dai et al. (2011) found that the radiosonde humidity records in IGRA need homogenization for its trend analysis, because of ubiquitous and large discontinuities resulting from changes to instrumentation and observing practices.
In view of the inhomogeneity problem of the CMA data, any discussion for a subregion can be uncertain and risky. So, we calculated the wind speed changes averaged over each of nine regions of China to check the consistency of wind speed changes in China. Following Duan (2009), the nine regions of China including the TP were identified based on the rotated empirical orthogonal function analysis and climate features. The stations within each region are shown in different colors in Fig. 1. The variations of annual mean surface wind speed in the nine regions are provided in Fig. 2. It can be found that the surface wind speeds of all regions other than regions G and H have the similar changes. Particularly, the step change around the end of the 1960s is an outstanding phenomenon and seems like a spurious one. Hereafter, we confirmed a physical basis for the observed step change based on independent observations.
a. Comparison between IGRA 850-hPa and CMA surface wind speed
Because radiosonde data for 1973–90 are unavailable for most IGRA stations, 1964–72 is used when analyzing the wind speed step change around the end of the 1960s. The annual mean wind speed at 850 hPa averaged over 31 IGRA stations with continuous 1964–72 observations was compared with CMA surface wind speed averaged at the collocated stations (Fig. 3). They have similar changes and their correlation coefficient is 0.70 (P < 0.05). The spatial distribution of trends in upper-air wind speed derived from IGRA changed similarly with the ones in surface wind speed, as shown in section 4d. As the two datasets are independent, we tend to believe both the CMA data and the IGRA wind speed data are qualified for climate analysis and their trends are quantitatively reliable.
b. Comparison between ground–air temperature gradient and surface wind speed
The surface sensible heat flux is mainly determined by the ground–air temperature difference (Tg − Ta) and surface wind speed together. When sensible heat flux dominates the surface energy budget, locally, a high wind speed usually results in a low value of the temperature difference and vice versa. Therefore, opposite trends between the surface wind speed and the ground–air temperature gradient are expected. Figure 4 shows that this phenomenon of opposing trends is rather apparent in spring over the TP, when the sensible heat flux is strongest. The correlation coefficient of the two time series is −0.76 (P < 0.01). This is additional evidence that the observed wind speed change in Fig. 2 is very likely real (not instrumental), as the measurements of ground temperature Tg, air temperature Ta, and surface wind speed are all independent.
From the above cross-validations, the wind speed step change around the end of the 1960s and the decline afterward are likely real, although artifacts such as instrument updates and station migration may contaminate the wind data to some degree (Cao and Yang 2012).
a. Observed trends of annual and seasonal mean surface wind speed
Figure 5 shows the observed variations of annual mean surface wind speeds averaged over China and the TP. Over the past 50 yr, the linear trend in annual mean surface wind speed is −0.010 ± 0.001 m s−1 yr−1 (error bar for P < 0.05) for the whole of China and a smaller value of −0.006 ± 0.002 m s−1 yr−1 (error bar for P < 0.05) for the entire TP region. However, it can be seen that the surface wind speed over China did not persistently decline. There are two transitions in wind speeds, occurring at the beginning of the 1970s and the 2000s; particularly, surface wind speeds step changed to strong ones around the end of the 1960s. These wind speed transitions are more significant over the TP. Regarding the annual mean daily wind speed, the difference in surface wind speed between the average over 1960–68 and the average over 1969–74 is 0.150 m s−1 over China and 0.431 m s−1 over the TP. On average, the magnitude of the daily mean increasing trend during the first period (1960–73: referred to as P1) is 0.011 ± 0.005 m s−1 yr−1 (error bar for P < 0.05) for the whole of China and 0.040 ± 0.009 m s−1 yr−1 (error bar for P < 0.05) for the entire TP region and then weakened during the second period (1974–2001: referred to as P2) with a trend slope of −0.018 ± 0.001 m s−1 yr−1 (error bar for P < 0.05) for the whole of China and −0.023 ± 0.001 m s−1 yr−1 (error bar for P < 0.05) for the entire TP region. Previous studies mainly addressed the wind speed decline over the second period. During the third period (2002–09: referred to as P3), there is no significant trend in the nation-averaged annual mean daily wind speed (−0.002 ± 0.006 m s−1 yr−1) other than the one averaged over the entire TP region (0.027 ± 0.013 m s−1 yr−1). During P3, the positive trends in wind speeds are detected for 0200 and 0800 BST in the whole of China and for all times in the entire TP region. In the atmospheric boundary layer, the surface wind speed usually exhibits a clear diurnal variation, because convection in the daytime boundary layer enhances the vertical mixing of momentum. This phenomenon is particularly prominent over the TP region, where intense sensible heat flux in the daytime enhances the development of very deep boundary layers (e.g., Luo and Yanai 1984; Ma et al. 2005; Yang et al. 2004; Yeh and Gao 1979). However, no obvious diurnal variation exists in the surface wind speed trends during all the periods over the TP. Changes in the boundary layer, therefore, should not be a considerable cause of wind speed change, which was also pointed out by Vautard et al. (2010). This implies that the wind speed change may be beyond the boundary layer scale.
Similar to the annual mean surface winds, seasonal mean surface wind speeds had transitions at the beginning of the 1970s and the 2000s (not shown). All the four seasonal mean surface winds consistently step changed to strong ones during P1 and weakened during P2 significantly. During both the strengthening period and the weakening period, the magnitudes of change for the seasonal mean surface wind speeds over the TP are larger than in the whole of China. A positive wind speed trend in winter [December–February (DJF)] and negative ones in other seasons are found in the whole of China during P3, but none of them is statistically significant (Fig. 6). However, the trends in all the seasons other than the spring are significantly positive during P3 over the TP region.
In a word, during the last 50 yr, there are three periods of wind speed changes in China. The trends over the TP are similar to the averaged ones over China but have larger magnitudes. In particular, the recent recovery (i.e., positive trend) of wind speed over the TP was found here, a phenomenon not previously reported in the literature.
b. Spatial distribution of observed trends in surface wind speed
Since the changes in surface wind speed over the whole of China and the TP are somewhat different, their spatial distribution is examined. Figure 7 shows the spatial distribution of the CMA surface wind speed trends during the three periods (P1, P2, and P3) defined in section 4a. Almost all the CMA stations in China exhibited positive trends in surface wind speed during P1, especially over the TP, where surface wind speeds increased considerably during this period (Fig. 7a). The averaged trend value is 0.010 m s−1 yr−1 over the whole of China versus 0.042 m s−1 yr−1 over the entire TP region. During P2, negative trends are dominant, with values of −0.018 m s−1 yr−1 over the whole of China and −0.024 m s−1 yr−1 over the TP (Fig. 7b). The spatial distribution of surface wind speed trends is complex during P3 as shown in Fig. 7c, perhaps because of too short of a period. Nevertheless, surface wind speeds at a large number of stations, especially over the TP, turned to increase during P3, although the wind speeds at the remaining stations still decline. The average value during P3 was −0.003 m s−1 yr−1 over the whole of China versus 0.022 m s−1 yr−1 over the TP.
c. Elevation dependence of observed trends in surface wind speed
From the above analyses, changes in surface wind speed over the TP are more apparent than those over the other regions of China. Previous studies (Liu and Chen 2000; Qin et al. 2009) have suggested that the warming rate over the TP is elevation dependent; we therefore examined whether the trends in surface wind speed are elevation dependent.
Figure 8 shows the elevation dependence of surface wind speed trend during the three periods defined before. All the P values of regressions during the three periods are less than 0.001. During P1, the positive slope value of 0.018 m s−1 yr−1 km−1 (0.7% yr−1 km−1) and the near-zero intercept of −0.007 m s−1 yr−1 (0.2% a−1) mean that surface wind speeds increased more drastically at higher elevations than lower elevations (Figs. 8a,b). In contrast, during P2 the negative slope value of −0.004 m s−1 yr−1 km−1 (−0.2% yr−1 km−1) and the negative intercept of −0.014 m s−1 yr−1 (−0.6% yr−1) indicate that surface winds declined more intensely at higher elevations than lower elevations (Figs. 8c,d). Figures 8e and 8f illustrate that surface winds during P3 strengthened more at high-elevation stations and weakened more at low-elevation stations, according to the negative intercept of −0.014 m s−1 yr−1 (−0.4% yr−1) and the positive slope of 0.010 m s−1 yr−1 km−1 (0.4% yr−1 km−1) of the fitted line. These results suggest that wind speed transitions occur earlier in high elevations than in low elevations.
Based on the above analysis, changes in surface wind speeds are more significant at higher elevations than at lower elevations. This result is consistent with the finding of McVicar et al. (2010). Surface winds at higher elevations are more sensitive to the changes in the upper-air winds, when the momentum is transported from the free atmosphere to the atmospheric boundary layer. Such elevation dependence of surface wind speed trends implies coincident changes in upper-air winds.
d. Trends in IGRA rawinsonde upper-air wind speed
At each rawinsonde station, trends in mean upper-air wind speeds at mandatory pressure levels of 850, 700, 500, 400, 300, and 200 hPa are derived from the IGRA rawinsonde dataset for the three periods. Figure 9 shows the spatial distribution of wind speed trends at 500 and 850 hPa, which are chosen to represent the lower free atmosphere over the TP and the remaining part of China, respectively. During P1, few (only two) stations in the TP region are available. Nevertheless, the upper-air winds speeds exhibited significantly positive trend values at all the selected levels. During P2, the negative trends in upper-air wind speeds prevailed under the level of 300 hPa and are significant over the TP at 500 hPa. During P3, positive trends in upper-air wind speeds in north China and the TP, and the negative trends in south China are detected from upper-air wind speeds above the level of 500 hPa, while the trends at 700 and 850 hPa are more complex.
Therefore, changes in upper-air wind speeds closely resemble to those in surface wind speeds during the P1 and P2. This resemblance is particularly distinctive between changes in surface wind speeds and those in winds at 500 hPa over the TP and between those in surface wind speeds and those at 850 hPa over non-TP. Although the spatial distribution of trends in upper-air wind speeds is not in complete agreement with surface wind speed trends during P3, the recovery of upper-air wind speeds is observed over the TP and some non-TP regions and a similar phenomenon also occurs in surface wind speeds. Since there is an intimate link between the surface and upper-air winds, changes in atmospheric circulation may exert more influence on changes in surface wind speeds than local changes in surface roughness, especially for the TP.
e. Trends in latitudinal gradients of 500-hPa geopotential height and surface temperature
Changes in geopotential height gradient are investigated to examine the intimate link between upper-air and surface winds. Figure 10a shows the temporal variation of geopotential height difference between the two latitude bands defined above (LL and HL) over the past 50 yr. The geopotential height gradient strengthened during P1, weakened during P2, and was enhanced again during P3. Such variations are synchronous with the observed upper-air and surface wind speed changes. It qualitatively confirms the coincident changes in regional atmospheric circulation.
The role of regional surface warming in the changes of wind speed under the background of global warming is further explored. Figure 10b shows the difference of surface temperature between the two latitude bands defined above. Clearly, the interannual variability of the surface temperature gradient is consistent with the geopotential height gradient at 500 hPa (Fig. 10a). Their correlation coefficient is as high as 0.88.
The causes of the wind speed changes are currently not fully understood. Several hypotheses have been proposed to explain the decline trends in surface wind speed, such as (i) changes in large-scale atmospheric circulation (e.g., Duan and Wu 2009; Guo et al. 2011; Jiang et al. 2010); (ii) increasing surface roughness (Vautard et al. 2010); (iii) effects of local air pollution (e.g., Jacobson and Kaufman 2006; Xu et al. 2006); (iv) influences of urbanization (e.g., Klink 1999; Li et al. 2011); (v) instrumental changes or observational drifts (e.g., DeGaetano 1998; Thomas and Swail 2011); and (vi) other possible causes given by McVicar et al. (2012). However, most of the factors may only explain the changes in surface wind speed in a local area and/or for a certain period. We demonstrated a reasonable explanation for the nation-wide surface wind speed changes in China over the last 50 yr.
The impact of instrumental changes or observational drifts cannot be ignored as it may yield spurious trends. Nevertheless, according the cross-validations in section 3, we illustrated that the trends derived from the observed data are not artifacts. The difference between the trends in urban and rural stations in China is quite small (Guo et al. 2011; Jiang et al. 2010), so urbanization should not be a considerable cause of surface wind speed changes. Direct anthropogenic impact over the TP is relatively small even though land-use change and industrialization have occurred over a small fraction of the TP. Air pollution, therefore, cannot be used to explain the observed wind speed changes and particularly their elevation dependence. Increasing surface roughness is regarded as an important factor, which attributes 20%–60% of Northern Hemispheric atmospheric stilling (Vautard et al. 2010). This may explain the contrasting trends between surface and upper-air wind speed over Europe and North America. However, this mechanism is not applicable to China, because the changes in upper-air wind speed coincided with the changes in surface wind speed according to the analyses made in section 3. Additionally, bare soils and short grasses are the dominant land-cover types and the surface roughness lengths are generally small (millimeters to centimeters) over the TP. The changes in surface roughness are quite insignificant as trends of NDVI for the TP during 1982–99 are shown to be very small in Beck et al. (2011). Thus, it is unlikely that the changes of roughness length affect the surface wind speed so significantly over the plateau.
As shown in the Fig. 10a, the three phases of surface and upper-air wind speed changes over China correspond well to the changes in the pressure gradient force and the atmospheric circulation. This substantiates the explanation that the changes of surface wind speed are largely induced by the changes in atmospheric circulation (e.g., Duan and Wu 2009; Guo et al. 2011; Jiang et al. 2010; You et al. 2010; Zhang et al. 2009b). Using a multimodel simulation during 1979–2000, Duan and Wu (2009) discussed the relationship between the East Asian subtropical westerly jet and latitudinal temperature gradient at 500 hPa. In this study, the role of regional surface warming in the changes of wind speed under the background of global warming was further explored (section 4e). The high correlation between the gradient of NCEP–NCAR 500-hPa geopotential height and the gradient of CRU surface temperature gradient is not an accident but has a physical basis. The regional difference in the surface warming rate may modify the gradient of geopotential height through thermal adaption, and this change in free atmospheric circulation will further influence the surface wind speed change through the momentum downward transport from free atmosphere to the atmospheric boundary layer. Therefore, the trends of the latitudinal gradient in surface temperature during the three periods approximately correspond to the ones of surface wind speed over China (see Fig. 5).
Several climate indices are used to characterize the atmospheric circulation variability. Therefore, it is worth discussing whether the wind speed change is related to the indices. As indicated in previous studies (e.g., Gong and Ho 2003; Gong et al. 2001; Ogi et al. 2003), AO/NAO variability exerts different impacts on the zonal mean wind speed at different latitudes of the Northern Hemisphere. However, both the AO and NAO indices have increasing (decreasing) trends before (after) 1990 (Figs. 11a,b), and these trends are not in phase with the changes in wind speeds over China. The PDO can modulate the impact of ENSO on the East Asian atmospheric circulation (e.g., Wang et al. 2008; Yoon and Yeh 2010), but we found no robust correlation between the PDO index and changes in wind speeds over China (Fig. 11c). Therefore, wind speed changes over China cannot be simply linked with oscillation indices such as AO/NAO/PDO. Nevertheless, we cannot exclude other climate indices or circulation patterns that are forcing the wind speed changes.
Based on temporal and spatial analysis of surface wind speed using the CMA operational meteorological data, we found that the surface wind speeds in China step changed to strong ones around the end of the 1960s. The winds then experienced weakening trend until the beginning of the 2000s and a possible recovery afterward. The high-elevation areas such as the Tibetan Plateau experienced more significant change during both the step change and decline period. Moreover, synchronous changes in surface wind and upper-air wind speed were observed over both the TP and the whole of China.
The deceleration of wind speed over the Tibetan Plateau since the 1970s was ascribed to the substantial tropospheric warming in the middle and high latitudes to the north of the plateau in two recent studies (e.g., Duan and Wu 2009; Zhang et al. 2009b). Our investigation was extended for a longer period and a larger region and supported the basic idea of the previous studies, but note that here we presented more direct evidence that associated wind speed changes with the latitudinal gradient of the surface temperature (instead of the 500 hPa in the two earlier studies), which may be easily measured. It is shown that there is a high correlation coefficient (0.88) between the latitudinal gradient of the surface temperature and the latitudinal gradient of geopotential height at 500 hPa in East Asia. The trend of the latter corresponds well to both the weakening and strengthening of wind speed over the Tibetan Plateau and over the whole of China. Therefore, the spatial inhomogeneity of global warming or cooling may significantly change surface wind speeds regionally via thermal adaption. This result contrasts with those for North America and Europe, where opposite changes in surface wind and upper-air wind speed were found and such changes can be partly explained by the increase of surface roughness lengths (Vautard et al. 2010). McVicar et al. (2012) note that the different causes of wind speed trends maybe occurred at different locations for different periods, and hence global attribution remains a key scientific challenge.
The intense variability of wind speed over the Tibetan Plateau is another indicator of the high sensitivity of the Tibetan Plateau responding to climate change. Therefore, the recovery of wind speed over the TP since 2002 might imply another transition of the regional atmospheric circulation and be a precursor of the reversal of wind speed trend over China, noting that both the wind speed and geopotential height gradient at 500 hPa over China turned to be increasing during recent years.
Finally, we noticed that the sudden increase of wind speeds both at the surface and upper troposphere during 1969–74 and 2002–09 generally coincided with periods of strongly negative AO phase. It is necessary to investigate whether equatorward displacement of jets occurred over Asian sector and, if so, how such anomalous would contribute to the increase of winds during those two periods.
This work was supported by the Global Change Program of Ministry of Science and Technology of China (2010CB951703), the Strategic Priority Research Program of the Chinese Academy of Sciences (Grant XDB03030300), and National Natural Science Foundation of China (Grant 41190083). The surface station data used in this study were provided by the Climate Data Center at the CMA National Meteorological Information Center. The involvement of Rong Fu in this work is supported by the Center of Earth System Science at Tsinghua University. We are very grateful to the anonymous reviewers for their valuable comments and constructive suggestions.