Wind speed trends over China: quantifying the magnitude and assessing causality

Authors


ABSTRACT

Temporal trends (1971–2007) in 10-m wind speeds from homogeneous observational data sets from 540 weather stations and reanalysis data sets are quantified and compared. Also, possible physical cause of inconsistencies between the data sets and temporal trends and variability in wind speeds are investigated. Annual mean wind speeds from the observational data exhibit pronounced downward trends especially in the upper percentiles and during spring. The NCEP/NCAR reanalysis reproduces the observed wind speeds, seasonality and temporal trends better than the ERA-40 even though it shows larger interannual fluctuations. The warm and cold AO and ENSO phases have significant influence on probability distribution of wind speeds, thus internal climate variability is a major source of both interannual and long-term variability.

1. Introduction

1.1. Temporal trends in near-surface wind speeds

A number of recent studies have reported declines in near-surface observed wind speeds during the past 30–50 years over parts of North America (Klink, 1999; Tuller, 2004; Pryor et al., 2009; Pryor and Ledolter, 2010), China (Xu et al., 2006; Jiang et al., 2010), regions of Europe (Pirazzoli and Tomasin, 2003; Brazdil et al., 2009) and Australia (McVicar et al., 2008). However, converse trends toward increasing wind speeds have been reported over the global oceans (Young et al., 2011). For example, based on data from passive microwave satellites, Wentz et al. (2007) found that wind speeds averaged over the tropics (30°S–30°N) increased by 0.04 m s–1/decade (0.6%/decade) over the period 1987–2006, while over all oceans the average trend was +0.08 m s–1/decade (1.0%/decade).

Temporal trends derived based on observed near-surface wind speeds are not always consistent with tendencies manifest either in other observational data or in reanalysis data sets. For example, in the analysis of daily mean wind speed data over Australia, measurements from terrestrial anemometers showed declines (a ‘stilling’) of −0.13 m s–1/decade when averaged over the entire country, but two gridded wind speed datasets (including the NCEP/NCAR reanalysis output) did not exhibit temporal tendencies (McVicar et al., 2008). Further, comparisons of 10-m wind speeds from observational data sets, reanalysis products and Regional Climate Model (RCM) simulations over North America showed trends in reanalysis data sets and RCM output were generally of lesser magnitude, and frequently of opposite sign, to those manifest in observational data sets (Pryor et al., 2009). Smits et al. (2005) also reported that the apparent decrease in storminess over the Netherlands based on station data was inconsistent with that based on reanalysis data, which suggested increased storminess during the same 41-years period.

The lack of correspondence between observational and reanalysis data sets and between observational data sets may derive from differences in time-series duration (and the short time-series of wind speed that are available for analysis), changes in station location, measurement height, data recording procedures and instrumentation deployed, in addition to difficulties in developing homogeneous observed records of near-surface wind speed (Pryor et al., 2009; Fu et al., 2011; Wan et al., 2010). Further, mid-latitude wind speeds exhibited high variability at interannual to interdecadal time scales linked to seasonal variation of the atmospheric circulation and internal modes of climate variability (Enloe et al., 2004; Pryor et al., 2005, Park et al., 2010; Pryor and Ledolter, 2010) which further confounds identification of robust trends and attribution thereof.

The discrepancies between temporal trends derived from observational data sets and from reanalysis or RCM simulations and over land surface areas and the oceans may also reflect, at least in part, a physical cause of the trends. Specifically, changes in surface roughness over the land surfaces have led to an increase in surface drag and a reduction in near-surface wind speeds. For example, model simulations over Eurasia using MM5 suggested that the recent increase in surface roughness (due to land-over change) explained 25–60% of the reported decline in 10-m wind speeds (Vautard et al., 2010). Since land-surface characteristics are not variable with year in the reanalysis data sets, changes in roughness length would not be characterized by the boundary data sets used within the reanalysis systems (Pryor et al., 2009). This mechanism, if confirmed, would also account for the discrepancy between temporal trends over land and water surfaces.

1.2. Atmospheric circulation and wind speed trends over China

The seasonal and interannual variability of Chinese climate is largely due to the summer and winter monsoons (Ding, 1994), which are major components of the global large-scale circulation and is linked to a number of internal climate modes as manifest in several teleconnection indices. For example, El Niño–Southern Oscillation (ENSO) is an important factor in the East Asian monsoon variability (Wu et al., 2003; Lim and Kim, 2007; Zhou and Wu, 2010). Further, Gong et al. (2001) found that the Arctic oscillation (AO) influenced the East Asian winter monsoon through the Siberian High. Later, Gong and Ho (2003) indicated that the AO significantly impacted the year-to-year variations in the East Asian summer monsoon rainfall via changes in the large-scale atmospheric circulation patterns. However, these prior studies have focused on linking the phase and magnitude of these teleconnection indices to variability in temperature and/or precipitation. Here we analyze relationships between intra- and interannual wind speeds and indices of the AO and ENSO.

Prior analyses of in situ daily average wind speed data from China have indicated declining values over the last few decades (Table 1). For example, Guo et al. (2010) estimated the spatially averaged trend in annual mean observed wind speed over China from 1969 to 2005 to be −0.18 m s–1/decade. Xu et al. (2006) reported a spatially averaged decline in annual mean wind speed (again computed from daily mean measurements) between 1969 and 2000 of 0.18–0.20 m s–1/decade. Using a similar data set, Jiang et al. (2010) reported a decline between 1956 and 2004 of 0.124 m s–1/decade, and most recently Fu et al. (2011) found an average decline in annual mean wind speed of 0.13 m s–1/decade between 1961 and 2007. A further study reported that over eastern Asia the annual mean surface wind speed changed by approximately −0.12 m s–1/decade between 1979 and 2008 (Vautard et al., 2010). Here we extend these prior analyses by specifically comparing temporal trends and seasonality as derived from the in situ data with those from global reanalysis data sets. We further seek to identify causality for any discrepancies identified and temporal trends and interannual variability.

Table 1. Synthesis of spatially averaged wind speed trend magnitudes (expressed in m s–1/decade) over China from previous studies of in situ data records. The trends are computed from annual mean wind speeds spatially averaged across China
Original paperTime-series durationNumber of stationsTrend (m s–1/decade)
Cong et al. (2009)1956–2005317−0.11

Jiang et al. (2010)

1956–2004535−0.12
Yin et al. (2010)1961–2008595−0.09
Fu et al. (2011)1961–2007597−0.13
Guo et al. (2010)1969–2005652−0.18
Xu et al. (2006)1969–2000305−0.2

2. Objectives

Herein, we present analyses of 10 m daily mean wind speeds at observational sites across China and output from reanalysis data sets to:

  • Examine the temporal evolution of mean near-surface observed wind speeds. In order to avoid ‘artificial’ trends caused by inhomogeneities within the data series, we also perform a homogeneity test on each measured wind speed time series to ensure consistency. We further examine the temporal trend in terms of consistency of the tendency over subsets of the entire record.
  • Quantify and compare the magnitude of wind speeds and temporal trends therein derived from observations and reanalysis products, and determine the (in)consistency among different datasets. We also evaluate the ability of the reanalysis data sets to capture the seasonality of wind speeds over China.
  • Further diagnose the dynamical causes of variability and trends. Specifically, we examine tendencies in reanalysis products at 850 hPa, and compare those with trends in 10-m wind speeds. We also examine the degree to which variability in near-surface wind speeds is attributable to variations in the atmospheric circulation as manifest in teleconnection indices.

3. Data and methods

3.1. Data sets

Data sets of daily mean wind speed and air pressure for 1971–2007 at multiple sites across China were obtained from the National Meteorological Center of China Meteorological Administration. These data meet the World Meteorological Organization's standards and all wind speeds were measured at 10 m above the ground. Even though most of these Chinese weather stations were established in the 1950s, we exclude data collected prior to 1971 due to the poor data recovery rates in the early decades and the countrywide introduction of new anemometers around 1970 (Xu et al., 2006; Jiang et al., 2010). For every station, a further data quality constraint is applied: that more than 85% observations are valid in each climatological season of each year.

Monthly near-surface (10 m) and 850 hPa u and v components of the wind speed from the National Center for Environmental Prediction/National Center for Atmospheric Research reanalysis data set (NCEP/NCAR, Kalnay et al., 1996) and European Center for Medium-Range Weather Forecasts 40 years Re-Analysis (ERA-40, Uppala et al., 2005) are also analyzed. Sea-level pressure fields and geopotential height at 850 hPa are also used to calculate the geostrophic wind speeds. The two reanalysis systems characterize surface fields at a spatial resolution 2.5 × 2.5°, so there are a total of 160 grids in mainland China. The wind components at 10 m from the NCEP/NCAR reanalysis data sets are designated as B variables because ‘they are partially defined by the observations but are also strongly influenced by the model characteristics.’ (Kalnay et al., 1996, p. 453) Observed land-based surface winds are not among the variables assimilated by the NCEP/NCAR reanalysis system and thus they are independent from the observational data analyzed herein. Near-surface winds from the NCEP/NCAR data set are derived from downward extrapolation from the lowest model layer using Monin-Obukhov similarity with seasonally varying surface roughness. Ten-meter wind components from the ERA-40 data set are also derived from downward extrapolation from winds at approximately 75 m height using Monin-Obukhov similarity with constant surface roughness.

Monthly mean values of the AO index and the ENSO index used in this study were obtained from the Climate Prediction Center (http://www.cpc.ncep.noaa.gov/products/precip/CWlink/MJO/climwx.shtml). The AO index is derived from empirical orthogonal function analysis of 1000 mb height anomalies in the NCEP/NCAR reanalysis product. The ENSO index is 3-month running mean of sea surface temperature (SST) anomalies (relative to a base period of 1971–2000) in the Niño 3.4 region (5°S–5°N, 120°W–170°W) calculated using Version 3b of the extended reconstructed SST dataset.

3.2. Methods

Observational records of wind speed are subject to large inhomogeneities due to site relocation, changes in anemometer, degradation of anemometer performance due to insufficient maintenance, changes in the observational environment and so forth. Thus here we examine each individual data time series of wind speed for inhomogeneities. The approach adopted is based on two methods: comparison of observed wind speeds with the geostrophic wind speeds computed from pressure gradients (Schmidt and von Storch, 1993; WASA group, 1998; Kristensen and Jensen, 1999; Tuller, 2004) and time-series analysis of the skewness of the wind speed data. First, we establish a series Ri of ratio of observed wind speed and geostrophic wind speed computed as shown below from mean sea-level pressure data in the NCEP/NCAR reanalysis:

display math(1)

where Vi and Gi are the annual observed wind speed and calculated geostrophic wind speed for each station in year i; math formula and math formula are the average of Vi and Gi from 1971 to 2007. The geostrophic wind speed is given by

display math(2)

with components:

display math(3)

where ρ is the density of air and f is the Coriolis parameter.

According to Kristensen and Jensen (1999), each pressure triangle determines unique constants a, b and c in the following set of equations:

display math(4)

where P1, P2 and P3 denote the three sea-level pressure values at three grid cells; X1, X2, X3 and Y1, Y2, Y3 are the coordinate value of the three grid cells, relative to the station position along the longitudinal and latitudinal direction, respectively. A shift in the time series Ri indicates a change in the relationship between the geostrophic and observed wind speed and thus is interpreted herein as indicative of an inhomogeneity in the observational record. In addition to Ri, another time series, the annual skewness of wind speed probability distribution at each station was computed. We postulate that any changes in, for example, anemometer sensitivity will be manifest in a change in the data skewness, because it would preferentially impact the ability to resolve very high and very low wind speeds (Pryor et al., 2009). Following Ebdon (1985), the skewness is calculated as:

display math(5)

where Vi is the observed wind speed in year i and Vik is the daily wind speed in year i.

A time series from a given site is identified as exhibiting an inhomogeneity based on the presence of breakpoints in time series of 7-year running means of the annual Ri and Si using the following subjective criterion:

  • If math formula, then the test series has a potential break at year i + 4,
  • If math formula, then the test series has a potential break at year i + 4, where
display math(6)
display math(7)

If the two test series exhibit breakpoints in the same year, the station was excluded from further analysis. Based on this data screening criteria, a set of the original data containing time series for 1971–2007 from 540 stations records were included in this study (Figure 1 shows the location of the stations). It should be noted that the data homogenization procedure applied here may exclude not only those stations that have inhomogeneities due to instrumentation changes or station moves but may also exclude stations subject to major (and relatively abrupt) changes in local land cover.

Figure 1.

Map of the 540 surface observing stations from which data are analyzed herein. Also shown are the major geographic features for the study region.

To quantify temporal trends in wind speed data series, regression analysis is applied to time series of annual mean wind speeds (either from a station or gird-cell) using a bootstrapping technique to find evidence for statistically significant trends at a significant level α = 0.05 (Pryor et al., 2009).

Because there is a strong seasonality in wind speed variations in China (Fu et al., 2011; Guo et al., 2010) and seasonality is an important element to measure climate, and further can be used as a diagnostic of the validity of the reanalysis data products. Here we use the seasonality index (SI) proposed by Walsh and Lawler (1981) to quantify the wind speeds seasonality in China:

display math(8)

where math formula is the average wind speed in a year and Vi is the monthly mean wind speed for month i. According to the Equation (8), the SI is a metric of monthly deviation from the overall mean value, and therefore it can be applied to quantify the seasonality in any parameter including wind speed.

The Kolmogorov–Smirnov (K–S) test (Ebdon, 1985) is a nonparametric technique used herein to determine the dependence of wind speeds on metrics of internal climate variability. Thus the K–S test is applied for each station to determine whether the wind speed probability distribution differ under conditions of high positive- or negative-phase climate modes (i.e. AO or ENSO index > |1|). In this analysis an extreme warm (cold) phase of teleconnection index is defined when the value of index is greater than 1 (or less than −1). For this work, the significance level at which the distributions are considered statistically significant different is α = 0.05.

4. Results and discussion

4.1. Evolution of wind speed time series from observational data

The rate of change in observed annual wind speed averaged over all the 540 stations in China from 1971 to 2007 is −0.17 m s–1/decade (Figure 2(a)). This trend is consistent in sign to earlier work summarized in Section 'Atmospheric circulation and wind speed trends over China' (Table 1), but as in that earlier work emphasizes the importance of the precise data period and data selection criteria in dictating the magnitude of the trend. The amplification in trend magnitude in analyses such as that presented herein which include the late 1960s and early 1970s emphasizes the influence of nationwide changes of anemometers (during this period), when there was a sharp increase in annual wind speed (Fu et al., 2011; Jiang et al., 2010). The ERA-40 reanalysis output does not exhibit a significant trend in 10-m wind speeds over China from 1971 to 2001, while spatial average NCEP/NCAR reanalysis wind speed exhibits negative trends at the rate of −0.13 m s–1/decade over the past 37 years (Figure 2(a)). In addition, as discussed by Guo et al. (2010), the temporal evolution of annual mean wind speed shows two phases: robust declines in the 1970s and 1980s, followed by a period of much smaller trends after 1990 (Figure 2(b)). Consistent with prior analyses of station data (Guo et al., 2010; Jiang et al., 2010), time series of the annual wind speed percentiles indicate that the downward trend in the upper percentiles is more pronounced than in lower percentiles. The 90th and 95th percentile wind speeds exhibit spatially averaged trends of −0.39 m s–1/decade and −0.5 m s–1/decade for 1971–2007. The decline in upper percentiles of the probability distribution of wind speeds is also in accord with the downward trend in dust storm frequency in China over the past 50 years (Zhu et al., 2008).

Figure 2.

(a) Annual mean wind speed over China during 1971–2007 based on the in situ measurements from the 540 stations, and two reanalysis data sets: NCEP/NCAR and ERA-40. (b) Trend lines for the four distinct periods within the observational data. Note the scale on the vertical axis of frames (a) and (b) differ, and has been expanded in frame (b) to more clearly illustrate the variations in the trend magnitudes as computed from the in situ data.

Consistent with the monsoon progression, observed wind speeds over China exhibit a distinct seasonal cycle, with higher wind speeds from late winter into spring, and lower values in August and September (Table 2). The spatially averaged temporal trends also exhibit seasonal variability. The largest magnitude decreases occur during the spring, while September which has the lowest mean wind speed also has the smallest spatially averaged decrease (−0.12 m s–1/decade) (Table 2).

Table 2. Monthly wind speed and trend magnitude during 1971–2007 averaged over all 540 stations across China
MonthJan.Feb.Mar.Apr.MayJun.Jul.Aug.Sep.Oct.Nov.Dec.
Wind speed (m s-1)2.292.522.82.982.82.522.352.192.182.232.322.24
Linear trend (m s-1/decade)−0.17−0.2−0.19−0.21−0.22−0.17−0.15−0.14−0.12−0.18−0.17−0.17

4.2. Intercomparison of wind speeds from in situ observations and reanalysis output

4.2.1. Mean wind speed

The ERA-40 reanalysis products end in the middle of 2002, so the intercomparison of mean wind speeds between the in situ observations and reanalysis data are conducted for 1971–2001. The average annual mean wind speeds from the two reanalysis data sets (NCEP/NCAR and ERA-40) and the observations exhibit a high degree of spatial similarity, indicating highest wind speeds in south coastal regions and northern China, from inner Mongolia to northeast (Figure 3(a)–(c)). Conversely, the central regions of China (e.g. the middle-reach of the Yangtze River) consistently exhibit relatively low wind speeds in all data sets. Histograms of the grid-cell averaged 10-m wind speeds from the reanalysis data sets and station observations reveal that the absolute magnitude of mean wind speeds calculated from the NCEP/NCAR reanalysis show greater agreement with the in situ observations than output from ERA-40, which exhibits negative bias (Figure 3(d)). The negative bias in 10-m wind speeds from ERA-40 relative to NCEP/NCAR output and/or observational records was also reported in the United States (Pryor et al., 2009) and Europe (Pryor et al., 2006). High magnitude wind speeds were also observed more frequently in the NCEP/NCAR reanalysis than in ERA-40 over the Arabian Sea and Bay of Bengal (Kumar and Philip, 2010). A prior intercomparison of 10-m wind speeds from the two reanalysis data sets over the Tibetan Plateau [which has an average elevation of over 4000 m, and is under-sampled in the in situ measurements (see Figure 1)] found ‘NCEP overestimates wind speed and ERA-40 underestimates it, with mean annual biases of +0.93 m s–1 for NCEP and −0.75 m s–1 for ERA-40’ (You et al., 2010). However, ERA-40 wind speeds at 10-m were found to be positively biased relative to observations in Australia (McVicar et al., 2008). The major discrepancies between 10-m wind speeds from ERA-40 and NCEP/NCAR are evident over the Tibetan Plateau. This finding combined with prior research over eastern China which has suggested ERA-40 better represents the mean sea-level pressure field than does NCEP/NCAR (Liu et al., 2012) may imply the biases in near-surface wind speeds in ERA-40 are linked to treatment of surface and topographic roughness, rather than the description of the atmospheric circulation.

Figure 3.

Spatial patterns of annual mean wind speeds for 1971–2001 from (a) in situ observational records, (b) NCEP/NCAR (c) ERA-40 using the scale shown in Figure 3(a) (unit: m s–1). Frame (d) shows histograms of the station and grid-cell average wind speeds from the three data sets.

As discussed in Sections 'Atmospheric circulation and wind speed trends over China' and 'Evolution of wind speed time series from observational data', there is strong seasonality of observed wind speeds in China, so seasonality is an important diagnostic of the reanalysis product skill. Because China is influenced by Asian winter and summer monsoons, the seasonality of wind speed in China is not only shown in wind direction but also in wind speed magnitude (Figure 4(d)). During winter, northwest winds prevail over most of China, especially the northeast. Conversely, southerly winds dominate over eastern China. Also, as shown in Table 2, winter wind speeds are generally higher than those during the summer. The SI computed for time series from each station and reanalysis grid-cell show large discrepancies. For example, the SI from the direct observational record is smaller in northeast and southeast China than in either of the reanalysis data sets, however, in the northwest corner, the SI exhibits higher similarity between observations and reanalysis products (Figure 4(a)–(c)).

Figure 4.

Average seasonality index for 1971–2001 from (a) observational records, (b) NCEP/NCAR (c) ERA-40 using the scale shown in (a). Frame (d) shows the mean wind vectors from the NCEP/NCAR reanalysis. The red vectors show conditions during the summer, while the blue vectors show the average during winter.

Histograms of the SI indicate that on average the grid-cell based SI calculated from the reanalysis data are positively biased relative to values computed from the in situ observations throughout most of China (Figure 5(a)). As shown by the spatially averaged monthly mean wind speeds in both the NCEP/NCAR and ERA-40 reanalysis data sets manifest wintertime maxima in wind speeds that are not indicated by the station observations (Figure 5(b)). This may be the main source of seasonality magnification in the two reanalysis data sets.

Figure 5.

(a) Histograms of the station and grid-cell average seasonality indices from the three data sets; observational records, NCEP/NCAR and ERA-40. (b) Spatially averaged mean monthly wind speeds over the period 1971–2001 from the three data sets.

4.2.2. Spatial distribution of temporal wind speed trends

When time series of the annual mean wind speed (derived from the daily mean wind speed) at each of the observational sites are subject to the bootstrapped trend analyses, 74% show a significant decreasing trend and 9% show a significant increasing trend. The proportion of stations exhibiting an upward trend over China is larger than that estimated by Fu et al. (2011) and Guo et al. (2010), but still represents only a small fraction of the total number of time series considered. Fifty-four percent of grid cells in the NCEP/NCAR product exhibit significant negative trends, while 3% show increases. Analysis of the ERA-40 output indicate 29% of grid cells with declining mean annual wind speeds and 16% show significant increasing trends (Figure 6). When expressed as a fractional change to remove the influence of the bias in wind speed magnitude described above, the trends are larger in the observations than in the reanalysis output (Figure 7(a)). This discrepancy may be due to reported changes in land use/land cover and thus an increase in surface roughness. Because land-surface characteristics used within the reanalysis products do not evolve (except seasonally) over the simulation period, changes in surface roughness would preferentially influence the in situ observations, potentially leading to larger temporal trends. However, because NCEP/NCAR output also exhibit large areas with negative tendencies, an additional important contribution to the observed reductions in annual mean wind speed derive from changes in the atmospheric circulation. In this context, it is important to note that a prior study has demonstrated large magnitude tendencies in the 500 hPa geopotential heights over the Tibetan Plateau over the period 1958–2001, but this tendency is not apparent in the ERA-40 output (Zhao and Fu, 2009).

Figure 6.

Temporal trends in wind speed data from (a) NCEP/NCAR and (b) ERA-40 overlain by station trends for the period of 1971–2001 (unit: %/yr). In each frame, a filled dot/a bold box indicates the wind speed trend at the station/grid cell is statistically significant at the 95% confidence level.

Figure 7.

(a) Histograms of wind speed trend magnitude (unit: %/yr) from the three data sets: observational records, NCEP/NCAR and ERA-40. (b) Comparison of trends computed as the average of trends from stations within each NCEP/NCAR grid cell and the value for that NCEP/NCAR grid cell.

It is worthy of note that when the mean trend for stations within each NCEP/NCAR grid cell are compared with the change computed from the NCEP/NCAR output (Figure 7(b)), there is a weak negative correlation. Thus, the discrepancies between temporal trends in station data and the reanalysis data sets are substantial in terms of the sign, spatial patterns and the absolute magnitudes. For example, as shown in Figure 6:

  • In northern China, the observational data indicate a dominance of negative trends, whereas grid cells from NCEP/NCAR and ERA-40 display increasing trends in the northeast and northwest, respectively.
  • In the regions around the middle reaches of Yangtze River, a substantial number of observational stations have time series that indicate a tendency towards increasing annual mean wind speed. This tendency is not apparent in the reanalysis data sets.
  • In the central and west-central areas of China, the NCEP/NCAR and observations exhibit generally negative trends, but the ERA-40 data indicate a positive tendency.

Inconsistencies between trends computed using different data sets has been reported in prior research focused on other geographic regions (see Section 'Introduction'). The causes for the discrepancies over China include, but are notlimited to change of observational instruments, siting and land-use/land-cover changes. An analysis designed to elucidate the physical causes of the temporal tendencies and interannual variability of wind speeds is given below.

4.3. Assessing the physical cause of near-surface wind speed trends and variability

A key question in interpreting the presence or absence of temporal trends is attribution of causality. Under the presumption (on the basis of analyses presented above) that the NCEP/NCAR reanalysis is a more skilful representation of the actual wind climate over China than ERA-40, the analysis of causality is focused principally (but not exclusively) on the NCEP/NCAR reanalysis product. As the first step in investigating possible causes of any temporal trends, the NCEP/NCAR reanalysis wind speeds at 850 hPa were also subject to trend analyses. The results suggest that the magnitude and spatial patterns of wind speed trends at 10 m and 850 hPa are consistent in terms of the spatial patterns and absolute magnitudes (cf. Figures 6(a) and 8). This coupled with the consistency between the sign of trends in NCEP/NCAR reanalysis and the in situ observations implies a primary cause of the trends over the period 1971–2007 is changes in the large-scale circulation.

Figure 8.

Temporal trends of wind speeds at 850 hPa from the NCEP/NCAR reanalysis computed for the period 1971–2001 (unit: %/yr). Values over topography higher than 850 hPa are masked out. A bold box indicates the trend in that grid cell is statistically significant at the 95% confidence level.

To further examine possible links to changes in atmospheric circulation, the mean sea-level pressure fields from the reanalysis products were used to compute a time series of spatially averaged annual mean geostrophic wind speeds over China which were also subject to a trend analysis (Figure 9(a)). The results indicate a very small magnitude downward trend in geostrophic wind speeds in the ERA-40 data set. This is consistent with the inferences draw above that changes in the large-scale atmospheric circulation are at least partly responsible for the declines in wind speeds over the period 1971–2001, and also the finding of small magnitude declines in ERA-40 10-m wind speeds (Figure 2). However, no significant tendency is apparent in geostrophic winds computed from the mean sea-level pressure fields from the NCEP/NCAR reanalysis (Figure 9(a)). This is in contrast to the weak downward trend in 10-m wind speeds from NCEP/NCAR, and may be due to inaccuracies in extrapolating to mean sea-level pressure over areas of complex topography, or it may reflect the presence of compensating positive and negative trends in different regions of China (see Figure 9(b)).

Figure 9.

(a) Annual mean geostrophic wind speed at sea level over China during 1971–2007 calculated from two reanalysis data sets: NCEP/NCAR and ERA-40. (b) Temporal trends of geostrophic wind speeds at 850 hPa from the NCEP/NCAR reanalysis computed for the period 1971–2001 (unit: %/yr). Values over topography higher than 850 hPa are masked out in (b).

The spatial patterns of trends in the geostrophic wind speeds computed from the 850 hPa height gradients in the NCEP/NCAR reanalysis (Figure 9(b)) are consistent with those derived from the 850 hPa wind components (Figure 8), and with the tendencies in 10-m wind speeds (Figure 6(a)), with declines over much of the south and west of the country and increases in the northeast.

To examine the role of major teleconnection indices on inter- and intraannual variability in wind speeds over China, the wind speed time series were conditionally sampled by the phase and magnitude of the AO and ENSO indices (Figure 10). For most parts of China except south of the Yangtze River, mean wind speeds are lower under the AO positive phase. This is consistent with previous research which has shown a positive correlation between wintertime temperatures and the AO index (Gong et al., 2001), and a priori expectations. Positive-phase AO is linked to a weakening of the Siberian High, leading to a reduction in the large-scale pressure gradient and thus a reduction in wind speeds. Application of the K–S test indicates that for more than half of stations, many of which are in northeastern China, wind speed probability distributions exhibit significant differences in the warm and cold phases of AO index (Figure 10(a)). The maximum divergence between cumulative probability distributions under positive and negative phases of the AO is between 30th and 70th percentiles, whereas the AO index appears to have relatively little impact on the distribution tail and thus extreme wind speeds.

Figure 10.

Difference in mean wind speeds under warm and cold (a) AO index; (b) ENSO index. Blue means negative value (i.e. lower wind speed under the positive phase of the index) and red means positive value. Filled circles indicate that the wind speed probability distributions are significant different under the warm and cold phases of AO and ENSO, according to the K–S test.

As a further test of the importance of the AO in determining wind speeds across China, the time series of spatially averaged observed wind speeds was subject to a nine-point Gaussian filter to remove the inter-decadal component and the residuals were correlated with the AO index. The correlation between the two time series is −0.47, confirming the assertion made above that wind speeds are generally suppressed under high positive-phase AO.

The data were also conditionally sampled by the phase and magnitude of the ENSO index (Figure 10(b)). The results indicate 55% stations exhibit lower mean wind speeds under positive ENSO index (warm-phase SST). According to the K–S test, the central portion of the wind speed probability distribution is also most strongly affected by the ENSO phase. Those stations which show significant differences in the wind speed probability distributions are primarily located south of the Yangtze River.

To further investigate the role of internal climate modes on inter-annual (and potentially inter/multi-decadal) variability in wind speeds, the spatially averaged wind speed data series were detrended using the fits shown in Figure 2(a), and the residuals computed (Figure 11(a)). These residuals represent the inter-annual variability after the long-term trend is removed and indicate the following:

  • On the whole, the variability of observed wind speed residuals approximates a U shape, with higher residuals in the 1970s and towards the end of the record (1990s and early twenty-first century) corresponding to the two phases of evolution of observed annual wind speed (Figure 2(b)).
  • Annual residuals from the NCEP/NCAR data set exhibit much greater variability than the manifest in the observations and ERA-40 data set. There is an abrupt change in the late 1970s in NCEP/NCAR reanalysis wind speeds variability, may be a response to the North Pacific regime shift that took place in the mid-to-late 1970s (Hare and Mantua, 2000). Further, variations in the data assimilation including the introduction of satellite data after 1978 (Kalnay et al., 1996) are another possible reason of discontinuities in the NCEP/NCAR reanalysis during this period.
  • The pattern of variability in annual residuals computed from ERA-40 reanalysis wind speed data sets exhibit greater agreement with the observations in terms of the magnitude of the residuals than the NCEP/NCAR reanalysis.
Figure 11.

(a) Detrended residuals in spatially averaged annual mean wind speeds from the three data sets: observational records, NCEP/NCAR and ERA-40. (b) Annual AO index and detrended variability in spatial average annual mean wind speeds from observations. The left vertical axis displays variability of wind speeds and the right shows the AO index.

Further analysis of the residuals computed from the observational data and annual AO index indicates that there is a strong negative relationship between them. In high AO periods the detrended residual tends to be negative, especially in 1989 when the AO index changed to a strong positive mode (Figure 11(b)), which is consistent with the above analysis that the wind speeds are lower under high positive AO index.

5. Summary and concluding remarks

Understanding the causes of temporal trends and variability in near-surface wind speed provides insights in climate dynamics and has relevance to impact studies of energy partitioning at the surface and thus agriculture, and dust storms and hence air quality. Here we analyzed data from in situ observations of daily average 10-m wind speeds from across China along with output from two reanalysis data sets in order to quantify temporal trends in the data sets, evaluate their consistency, examine causes of inconsistencies between the data sets and diagnosis causes of any temporal trends. Because time series of near-surface wind speeds from observational records are vulnerable to large inhomogeneities, here we adopt a new approach to select homogeneous dataset of wind speed in China and examine variations from 1971 to 2007.

Results of analyses presented herein indicate:

  1. Annual mean wind speeds computed from in situ measurements of daily mean wind speeds exhibit declining trends over much of the country during the period 1971–2007. The spatially averaged mean trend at 10-m wind speeds is −0.17 m s–1/decade. The trend is of larger magnitude in the upper percentiles of the wind speed probability distribution and during the spring months. This trend is of greater magnitude during the 1970s, and subsequently is almost equal to 0. It is possible that at least some fraction of the initial decline in wind speeds is attributable to a change in instrumentation that occurred early in the 1970s.
  2. Intercomparison of direct in situ observations and those from the reanalysis data sets indicates the patterns exhibit a high degree of spatial similarity, indicating highest wind speeds in south coastal regions and northern China and lowest wind speeds in the central regions. However, the NCEP/NCAR reanalysis reproduces the magnitude of the observations better, while data from the ERA-40 reanalysis are negatively biased. Both the two reanalysis products overestimate the observed wind speed in winter, leading to the magnification of seasonality.
  3. As in previous studies over the United States and Australia, the temporal trends between observational data and reanalysis data sets differ both in magnitude and spatial signatures. For example, 74% of time series of annual mean wind speed derived from station observations exhibit a significant decrease, but only 54 and 29% of grid-cell averaged data from NCEP/NCAR and ERA-40 reanalysis products exhibit decreasing wind speeds. Spatially averaged trends (1971–2001) from the two reanalysis data sets are smaller than those derived from observations. For example, the spatially averaged temporal trend in the NCEP/NCAR reanalysis data over the past 37 years is 77% of that from the station time series.
  4. Temporal trends from stations are averaged within each NCEP/NCAR grid cell are negatively correlated with trends computed from the NCEP/NCAR reanalysis, indicating the spatial patterns of temporal trends are not consistent between these data sets.
  5. The warm and cold AO and ENSO phases strongly impact the probability distribution of wind speeds. Under positive-phase AO, mean wind speeds are lower over most of China. Under positive ENSO phase, the mean wind speeds are higher in north of China and lower in south of China.
  6. Detrended time series indicate the annual residuals of wind speeds from ERA-40 exhibit greater accord with the observational time series, whereas data from the NCEP/NCAR reanalysis exhibits larger variability.
  7. The AO index has a strong negative influence on interannual variability of observational wind speeds. There has been a marked decline in sea ice coverage over the Arctic during recent decades (Lindsay and Zhang, 2005) which has been linked to shifts towards the positive phase of the AO (Holland, 2003). Thus, given the positive-phase AO is linked to lower wind speeds over China, the AO may be a contributing factor of both interannual variability and also the longer-term tendency reported herein. Further analyses to quantify causes of the tendencies reported herein is warranted.

Acknowledgements

The authors acknowledge financial support from the National Natural Science Foundation of China (#40875059), Special Funds for Scientific Research on Public Causes (No.GYHY201006038) and Innovation Project of Postgraduates Research in Jiangsu Province (CX10B_287Z). S. C. P. acknowledges support from the National Science Foundation (grant #1019603). The authors acknowledge the insightful comments provided by two anonymous reviewers.