Interannual variability of the summer wind energy over China: A comparison of multiple datasets

Key Laboratory of Meteorological Disaster of Ministry of Education, Collaborative Innovation Center on Forecast and Evaluation of Meteorological Disaster, Nanjing University of Information Science and Technology, Nanjing, China State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics (LASG), Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, China CAS Center for Excellence in Tibetan Plateau Earth Sciences, Chinese Academy of Sciences (CAS), Beijing, China University of Chinese Academy of Sciences, Beijing, China

The long-term changes in wind speed have been evaluated in previous studies. In the United States, the 10-m (10 metre above the ground) wind speed was found to have decreased significantly over the past 30-50 years according to data from approximately 1300 stations over the contiguous United States. 2 If the annual mean wind speeds are arranged in ascending order, more than half of the annual 50th percentile 10-m wind speed and nearly half of the 90th percentile annual 10-m wind speed time series exhibit significant decreasing trends. 8 Decreasing trends are also found in the annual mean surface wind speeds over Europe. 6,9,10 In addition, the daily maximum wind speed also shows decreasing trends over northwestern Europe in summer. 11 In China, the 10-m wind speeds derived from meteorological station data also exhibited a long-term decreasing trend from 1960 to 2010. [12][13][14] The projection of future changes in wind energy has been a focus of previous studies. Many of these kinds of studies are based on the analysis of global and regional climate models. For example, it is projected that there will be an increasing trend in wind energy resources in parts of wind farms that have large-scale generation capacity over the United States. Over northern Europe, the mean wind speeds, 90th percentile wind speeds, and energy density in the period from 2081 to 2100 are projected to be slightly lower than those during 1961-1990. 7 The annual and winter mean WPD in China during the early 21st century is also projected to decrease slightly. 15 In addition to long-term changes, the WPD also shows robust interannual variations. The year-by-year or interannual variability in WPD is important for the production of wind farms. 16 Wind farms usually face uncertainties in predicting the energy output due to the robust interannual variations in the WPD. 17 The year-to-year changes in WPD could cause substantial economic losses to wind farm companies. However, in comparison to future change projections, few studies have focused on the interannual variability in WPD. Pryor et al. 16 assessed the historical variability in annual wind indices across Europe and found that the North Atlantic Oscillation (NAO) contributed to the interannual variability of the 90th percentile winter wind speeds. Additionally, the interannual variability in WPD over Europe is also attributable to variations in the Atlantic Oscillation (AO). 18 The El Niño-Southern Oscillation (ENSO) is a coupled ocean-atmosphere cycle with a 2-to 7-year period occurring over the tropical Pacific Ocean. The interannual variability in the global climate is mainly affected by ENSO. 19 The sea surface temperature (SST) anomalies over the Niño-3.4 region (5 S-5 N, 120 -170 W) are used to track ENSO, which is also referred to as the ENSO index. [20][21][22] A positive ENSO index represents the warm phase (El Niño), whereas a negative ENSO index represents the cool phase (La Niña). El Niño also plays an important role in the year-toyear variations in wind energy. For example, over the Canadian Prairies, the surface wind speed stills during winter in El Niño years. 17 In the Great Lakes region, large interannual variability occurs in the winter and small variations occur in the summer, which are associated with El Niño events. 23 The wind energy over continental China also shows large interannual variations. In China, the total electricity consumption is high in the summer months and tends to increase with global warming. 24 East China is controlled by a large summer monsoon system, and the interannual variations in wind energy in summer are greatly affected by the East Asian summer monsoon. ENSO has been demonstrated to be a major factor that modulates the interannual variations in the East Asian summer monsoon. [25][26][27][28][29][30][31] Following the monsoon changes, the wind energy also exhibits corresponding changes. For example, more than 55% of the observational stations in China show weaker than normal near-surface wind speeds in El Niño years, 32 but the underlying mechanisms remain unknown. The major objective of this study is to reveal how El Niño modulates the interannual variations in summer mean WPD over China. The performances of different reanalysis data in quantifying the summer mean WPD changes over China are also assessed.
The remainder of the paper is organized as follows. Section 2 describes the datasets and analysis methods used in this study. In Section 3, the main results are presented, including the interannual variability of the WPD and the effect of El Niño. The concluding remarks are presented in Section 4.

| Data description
We use the daily mean 10-m wind speed from 756 meteorological stations as observation data. These data were obtained from the China Meteorological Administration (CMA; http://cdc.nmic.cn/home.do). The observations were quality controlled, including internal consistency checks, spatial and temporal consistency checks, identification of outliers, and correction of suspected and erroneous data, but homogenization was not conducted. 33 We chose 503 stations with the best continuity (i.e., the missing data represent less than 5% of the data in each year) during 1960-2013 in our analysis. The monthly mean data were calculated from the daily observation data.
In this study, six reanalysis datasets were used: (1) National Centers for Environmental Prediction (NCEP)-U.S. Department of Energy (DOE) Reanalysis 2 (NCEP-2), which is an improved version of NCEP-1; 34 (2) Modern-ERA Retrospective Analysis for Research and Applications (MERRA), which is a National Aeronautics and Space Administration (NASA) reanalysis for the satellite era using a major new version of the Goddard Earth Observing System Data Assimilation System Version 5 (GEOS-5); 35 (3) Japanese 55-year Reanalysis Project (JRA-55), which employs four-dimensional variational data assimilation (4DVAR) with variational bias correction for satellite radiances; 36 (4) Interim European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis product (ERA-Interim), which improves certain key aspects of ERA-40 and employs the 4DVAR assimilation method 37 ; (5) 20th Century Reanalysis (20CR), which is performed with the ensemble filter 38 ; and (6) SST from the Hadley Centre sea ice and sea surface temperature dataset (HadISST). 39 Detailed information on the six reanalysis datasets is presented in Table 1.
Monthly air temperature, monthly surface pressure, daily meridional wind, daily zonal wind, and monthly SST are used in this study. The monthly mean WPD is calculated from daily data in the reanalysis data.

| Homogenization of the daily 10-m wind speed data
Because the original meteorology station data (MSD) station data were not homogenized, we performed a homogenization analysis of the data before they were applied. The penalized maximum F test 40 is applied to perform the homogenization. This test can detect any discontinuities in the time series of the daily 10-m wind speed data. Each station is detected for all times. This method does not require a reference series or station because it allows a long-term trend and accounts for first-order autocorrelation. 33 We first used the penalized maximum F test to detect the change point in the time series at each station. If significant change points are identified, the quantile-matching method is used to remove the discontinuities in the daily 10-m wind speed around the detected change points. Then, the MSD is homogenized and used to calculate the monthly WPD over China. For details of the method, please refer to Dai et al. 41 and Wang et al. 42

| Computation of WPD at the turbine hub height
The WPD depends on the air density and wind speed. WPD is used to describe wind resources 43,44 and indicates how much wind energy can be harvested at a location. 45 The WPD at each time step can be written as where U is the near-surface wind speed and ρ is the air density, which is equal to 1.225 kg.m −3 in this study.
The turbine hub height is 80 metre above the ground in China. Therefore, we first need to calculate the 80-m wind speeds. The vertical wind profile is applied to extrapolate the wind speed from 10-m up to 80-m. The empirical power law, shown in Equation 2, was first used by Elliott 46 and has been widely used. 7,47,48 where U Z is the wind speed at the turbine hub height Z and U 10 is the wind speed at 10-m. This formula is often used in the surface layer chartered by neutral conditions and smooth areas without variations in the boundary layer stability. Therefore, Justus and Mikhail 49 where U Z is the wind speed at the turbine hub height Z and U 10 is the wind speed at 10-m. In this study, we first use the alternative formula to calculate the 80-m wind speed. Then, the WPD at 80-m is calculated by using Equation 1.
Before the area-weighted mean WPD was calculated, the station data were interpolated onto a 1.25 × 1.25 resolution grid using bilinear interpolation. Because the focus of our study is interannual variability, variations longer than 9 years were filtered out from the original datasets with a Lanczos filter 50 before the analysis. We focused on the time period from 1985 to 2004 in this study because of the availability of data.

| The climatology of the WPD
The 20-year averaged spatial pattern of the summer mean WPD is shown in Figure 1. The quality of the reanalysis datasets was quantitatively measured by the pattern correlation coefficient (PCC) and root mean square error (RMSE) with reference to the MSD, as given in Table 2. As listed in

| The interannual variability in WPD
To reveal the interannual variation in the summer mean WPD, the standard deviation was first calculated. The spatial distributions of the interannual variability are shown in Figure 2. To better understand the spatial and temporal distributions of the summer mean WPD in China, empirical orthogonal function (EOF) analysis was performed to identify the leading interannual variability modes. Because the focus of our study was interannual variability, variations longer than 9 years were filtered out from the original datasets with a Lanczos filter 50 before the EOF was calculated. The MSD results are shown in Figure 3. The first leading EOF mode (hereafter EOF1) ( Figure 3A) Figure 3B. The MSD exhibited significant interannual variations. The PC1 of MSD was regressed on the five reanalysis datasets, and we obtained the spatial patterns of the first leading interannual variability modes derived from the five reanalysis datasets (Figure 4). All reanalysis datasets could reproduce the observed characteristics of EOF1, with positive anomalies south of the Yangtze River and negative anomalies north of the Yangtze River. Only MERRA and ERA-Interim showed significant interannual variations in these two regions. The weakness of this analysis was that the positive and negative anomalies were stronger than the observed values. In NCEP-2, positive anomalies were also observed in the Tibetan Plateau.
In Figure 4, significant anomalies occurred in the south (22 N-30 N, 104 E-118 E) Yangtze River (gray box in Figure 4A). All reanalysis datasets reasonably reproduced that observed spatial pattern. Over the north of the Yangtze River region, there were also significant anomalies.
In the following analysis, we focus on these two specific regions, namely, north and south of the Yangtze River, where wind energy has developed rapidly in recent years. 52

| The effect of ENSO
To reveal the potential link of WPD variations with SST anomalies, we regressed PC1 of the summer mean WPD EOF1 with SST in Figure 5. Note that the El Niño mature phase occurs during boreal winter, namely, D (−1) JF (0). Here, −1 indicates 1 year prior to the summer mean WPD, and 0 indicates the current year. The SST anomaly features a typical El Niño pattern, with significant positive anomalies located in the equatorial central and eastern Pacific ( Figure 5A). The correlation coefficient between PC1 and the ENSO index, which is defined as the regional average of SST anomalies within the 5 S-5 N, 120 -170 W box, was 0.576, which is statistically significant at the 99% confidence level. Hence, ENSO is a major factor that can affect the interannual variations in the summer mean WPD.
The evolution of El Niño includes three stages: developing (boreal summer), peaking (winter), and decaying (following spring). Following the evolution of El Niño events, a warming tropical Indian Ocean (TIO) was observed. 53 TIO warming appeared in boreal winter ( Figure 5A) and reached a maximum during MAM (0) ( Figure 5B). In El Niño decaying year summers, the SST anomalies in the tropical eastern Pacific were generally neutral, whereas significant warming was evident in the TIO ( Figure 5C).
To quantify the impact of El Niño on the summer mean WPD, the lead-lag correlation of PC1 with the ENSO index and the area-averaged SSTA over the TIO (10 S-10 N, 40 E-110 E) region are shown in Figure 6. We denote the year with the summer mean WPD time series at year 0 and the leading (following) year is denoted as year −1 (1). Whereas the correlation coefficient of WPD PC1 with the ENSO index decayed rapidly from boreal winter to the following summer, the significant positive correlation of WPD PC1 with the TIO index persisted from boreal winter to the following summer. The highest correlation coefficient occurred in JJA (0). This kind of relationship is evident in both the observations ( Figure 6A) and the ERA-Interim reanalysis ( Figure 6B). The delayed response of the TIO is also evident in Figure 5C. contribute to the increase in the WPD in JJA (0) over this region. In the meantime, the wind over the north of the Yangtze River valley will be weakened.
The results from present studies show that the formation of the WNP anticyclone anomaly contributes to TIO warming from winter to spring, 54,55 and warming could persist into decaying summers. 56 The TIO warms similar to a battery charging capacitor. 56 After El Niño decay, TIO warming persists through JJA (0) and exerts its climatic influence on the surrounding regions. The dynamic 53 and thermodynamic 57 mechanisms are used to explain how TIO warming could affect the WNP anticyclone. The basin-wide IO warming does not affect the WNP circulation in the F I G U R E 7 Anomalies of 10-m wind (mÁs -1 Áa -1 ) regressed onto the El Niño-Southern Oscillation (ENSO) index (vector). The shaded areas indicate grid points where the anomalies are statistically significant at the 5% level. Zero indicates the year of the summer wind power density (WPD) time series, and −1 (1) indicates the leading (following) year [Colour figure can be viewed at wileyonlinelibrary.com] mature phase of El Niño 58 but is the major cause of WNP circulation during decaying summer. 59 The local cold SST anomalies during boreal winter and the subsequent seasons could maintain anomalous anticyclones. 28 Significant cold SST anomalies exist in the WNP ( Figure 5) during D (−1), JF (0), and MAM (0). Thus, the anomalous anticyclone in the WNP during the El Niño decaying summer may be affected by both basin-wide IO warming and local cold SST anomalies. 60 The basin-wide IO warming will induce anticyclonic anomalies through Kelvin wave-induced Ekman pumping divergence, 58 and the local cold SST anomalies will maintain the anticyclonic anomaly through the IO forcing effect. 61 More detailed information can be found in the study by Li et al. 62 In addition, we found that the near-surface wind speed is small during JJA (0) over the north of the Yangtze River ( Figure 7C). China's topography is varied and complicated, especially in northern and western China. The near-surface wind speed is influenced by the topography. Thus, we examined the relationship between the surface pressure over land and El Niño (Figure 8). Positive anomalies were observed over the low-pressure center (Tibetan Plateau), and negative anomalies existed over the high-pressure area (low-altitude region) during JJA (0) after El Niño events. This result means that the pressure gradient decreased north of the Yangtze River. The lower pressure gradient helped the wind decrease during the following summer. Therefore, a lower WPD exists during the following summer over the northern Yangtze River.

| SUMMARY
In this study, the interannual variability of the summer WPD derived from five reanalysis datasets (NCEP-2, MERRA, JRA-55, ERA-Interim, and 20CR) and observations are compared. The mechanism of the interannual variability is investigated. The major findings are summarized below.
1 The observed long-term mean summer WPD over China is large in northern and eastern China and small in the Sichuan Basin. All reanalysis datasets, except NCEP-2, reasonably reproduce the observed spatial distribution but include evident biases in intensity. Among the five reanalysis datasets, MERRA and ERA-Interim show the best performance with large PCC and small RMSE values. 3 The interannual variability of the summer mean WPD over China is dominated by ENSO. In the south of the Yangtze River, the WPD increases during El Niño decaying year summers but decreases during La Niña decaying year summers. In the north of the Yangtze River, the opposite change occurs. During the El Niño decaying summer, an anomalous anticyclone appears in the WN Pacific. This condition affects the continent south of the Yangtze River, and a strong anomalous south wind appears, leading to an increase in the WPD south of the Yangtze River. During the El Niño decaying summer, the near-surface wind speeds are slow north of the Yangtze River, which can be explained by the low surface pressure gradient.