Increases in China’s wind energy production from the recovery of wind speed since 2012

China has realized a 56-fold increase in installed wind capacity, from 5.9 GW in 2007 to 328 GW in 2021. In addition to increasing installed capacity, plans to substantially increase wind energy production for climate change mitigation also depend on future wind speeds, which strongly influences the efficiencies of installed turbines within individual wind farms. A reversal in globally decreasing wind speeds over several decades has been reported previously. However, subsequent studies using other data sources reported only a slight increase or no reversal in China. These uncertainties regarding China’s wind energy production hamper estimates of wind energy production potential. Here, our analysis of quality-controlled wind speed measurements from in-situ stations shows that the wind speed decline in China reversed significantly since 2012 (P < 0.001), but with substantial spatio-temporal variability. We further estimated the capacity factor (CF) growth and the wind power gain solely associated with the changes in wind speed ranges from 31.6 to 56.5 TWh yr−1 based on the 2019 installed capacity. This estimate explains 22.0%–39.3% of the rapid increase in wind generation CF in China during 2012–2019. The result implies that the site selection of wind farms should consider both current wind situation and future wind speed trends. Further studies are needed to understand the driving factor of wind speed recovery in support of the wind energy industry.


Introduction
China is currently the largest CO 2 emitter in the world [1], and has an ambitious plan for emissions to peak by 2030 and achieve carbon neutrality by 2060 [2]. A potentially efficient pathway for curbing CO 2 emissions is to replace fossil fuels with electric power and decrease the carbon intensity of electricity production through renewable energy, e.g. wind power [3]. The cumulative installed wind power capacity in China has grown exponentially from 5.9 GW in 2007 to 328 GW in 2021 [1,4,5]. With over one-third of the world's wind power capacity, China is the world leader in wind power generation [6,7]. Looking into the future, China's total installed capacity is anticipated to reach 800 GW by 2030 and 3000 GW by 2060 with an additional ∼70 GW installed annually [2].
The efficiency of wind turbines for generating electricity is described as the ratio of the generated power to the rated power, known as the capacity factor (CF) [8,9]. While the rated power is solely dependent on the number and type of turbines that are installed, the generated power additionally depends on wind farm management (e.g. control of wake effects, maintenance and repair), and wind speed changes [6,10,11]. The theoretical maximum wind power is proportional to the cube of wind speed [10]. The amount of those power that can be converted into electricity for any given turbine is further limited by the wind turbine power curve, which can be improved through technology innovation including optimization of blade aerodynamic structure and blade materials [11]. Once turbines have been installed, the power curves are fixed, and the key factor driving power generation is wind speed change [9][10][11]. Yet, wind speed's impact on the CF for installed wind turbines in China remains unknown, largely owing to (a) debate on how average wind speed has been changing during the past decade when most wind turbines were installed [12][13][14][15][16]; and (b) difficulty in translating the in-situ wind records (10 m above ground, daily and discontinuous) into wind power production (i.e. CF) [17][18][19].
As for wind speed over China, most studies [20][21][22][23] report a long-term decreasing trend during 1960-2010, which is consistent with the terrestrial stilling that has been reported globally [12][13][14]. However, there is no consensus on how wind speed in China has changed since 2010, which marks the beginning of the period when most (more than 85%) wind turbines were installed. Some studies have reported a continuation of the stilling [21][22][23][24][25]; others show a reversal [9,15]. These differences relate partially to the analysis of in-situ records from different subsets of meteorological stations and the application of different quality control criteria [9,16,[20][21][22][23]25] and lead to the influence of wind speed on wind energy production unknown. For the subsets, recently the HadISD and GSOD datasets have been found to record calm winds inaccurately, particularly for Asian stations including China, producing erroneously high wind speed in recent years [26]. Therefore, the results of Zeng et al [9] are likely biased for Asia. It is therefore urgent and necessary to use both China's official dataset and the updated global datasets to determine more accurate wind speeds in China. For the quality control method, Yang et al [12] and Tian et al [14] also used China's official dataset.
Their quality control method necessitated the exclusion of about half of the stations that changed locations over time, potentially biasing the observed wind speed trends, especially for the past decade. Herein, our research aims to clarify the wind speed trend in China using all the four datasets from previous studies and a viable quality control method that preserves more data records without compromising accuracy. Furthermore, we convert the wind speed change to wind energy CF change to quantify the effect of wind speed change on wind energy production.

Wind speed datasets
We used the China Surface Climatic Data Daily Data Set (CSD) (Version 3.0), from the China Meteorological Data Service Center (CMDC; http://data.cma. cn/en/?r=data/; last accessed March 2020). The CSD contains daily mean (average of four observations at 2:00, 8:00, 14:00 and 20:00 now and, three observations at 8:00, 14:00 and 20:00 before installing the automatic recording) data from 699 national-level reference climate stations and basic climate stations in China (not including stations in Hong Kong, Macao and Taiwan) from 1951 until three months before the download time (2020.03) (table S1). Data from 1951 to 2010 were checked for quality both manually and automatically [27]. To get a complete and reliable time series, we only used data from stations for which more than 15 daily observations were available for each month during 1990-2019 [9]. The 15 d criterion is a safety threshold with a sensitivity test, in which we tried 10, 15, 20 and 25 d as thresholds. Benefiting from the integrity of the data, the difference is limited as supplementary figure 1 shows. The distribution of the stations covers all of China's provinces but has a denser coverage in the South East compared with other regions (inset in figure 1). We also applied this quality control method to the other three datasets (see more in supplementary text).
For seasonal analysis, we defined spring as March, April and May, and winter as the following December, January and February, with summer and autumn accordingly defined. For regional analysis, we divided China into six regions based on the partitioned model used in the 2019 Wind Power Industry Report [28], shown in figure 2(a).

Piecewise regression model
We used the piecewise regression Model 1 mentioned by Toms & Lesperance (2003) to determine turning points and generate the trend lines for the wind speed time series [29]. Model input variables are wind speed time series and the earliest and latest plausible dates for the turning point. We set the turning point to be at least 2 years away from the start and end of the analysis period (1990-2019) to ensure there were sufficient data for robust testing. The output from the model is the timing of the turning point, the trends before and after the turning point, the R 2 for the piecewise regression, the P value for the timing of the turning point, the first and the second partial fit and the complete model fit. The calculation of the P value is based on the t-test null hypothesis: the trend difference before and after the tested point is 0.

CF calculation
We refer to three types CFs: (a) real CF (CF real , orange line shown in figure 3) is based on the data statistics; (b) calculated raw CF (CF raw , left axis in figure 4) is our raw calculation from wind speed; (c) calibrated CF (CF cali , right axis in figure 4) is CF raw calibrated by CF real as shown in equation (4).
Both CF real and CF raw are calculated from the following equation: For national CF real , the rated power (P rated ) is the national installed wind capacity (unit: GW) and the real power (P real ) is the annual net wind power generation (unit: GWh) in China taken from the U.S. Energy Information Administration (www.eia. gov/opendata/) divided by 8760 h (hours in a year). The provincial CF real s come from the annual wind power grid-connected operation statistics reported by National Energy Administration of China (www.nea. gov.cn/). National and provincial CF raw s were calculated from monthly average wind speeds as follows: (a) we converted monthly wind speeds data into hourly data; (b) we used a power curve to convert wind speed to output power, allowing the calculation of the CF; (c) we use Thiessen Polygon to calculate the national and provincial average CF raw s. For the first step, we used 33 stations with hourly observations in the corrected HadISD dataset as reference stations (the distribution of the reference stations is shown in figure S2a). The hourly wind speed distribution for a given month was assumed to follow the Weibull distribution associated with the nearest reference station (figure S3, equation (2)) [30,31]: where λ and k are the parameters that determine the distribution using the function called 'wblfit' in MATLAB. A detailed description of this method is provided by Sweerts et al [32].
For the second step, we used an exponential wind profile power law relationship [14] to convert the observed wind speeds (u at height z s = 10 m) to wind speeds at the hub height of the wind turbines (u tb at hub height z tb = 85 m): where α is estimated using equation (3)  . The GW 121/2500 turbine is selected as Goldwind has the largest wind power market share in China and this turbine is one of the most commonly installed [34]. With the energy production, we can calculate the CF raw s according to equation (1). For the third step, we calculate the areas of the polygons in the Thiessen Polygon map (figure S6) determined for the stations in CSD dataset. The calculated national and provincial CF raw s are reported as the area weighted average values for these polygons. The calculated CF raw s are lower than CF real s because we use weighted area rather than actual wind farm distribution. In reality, wind farms are only located in areas with good wind resources, and the type of wind turbine was always carefully selected to match local weather conditions. Using CF raw will underestimate the contribution of the average wind speed change to the CF real change. Therefore, when calculating the gain driven by wind speed recovery, we first converted CF raw into real magnitude. The multiplier (β = 1.78) was defined as the ratio of CF real (0.225) and CF raw (0.126) in 2019 (our turbine data was based on the technology in 2019). This way, the effect of technology improvement is controlled to be the same. The CF cali was calculated as where β is only suitable for a national value of 2.5 MW turbine and should not be used for provincial values nor for other turbines.

The recovery of wind speed in China since 2012
Our results show that after several decades of decreasing trend, the wind speed in China has significantly increased since 2012 (P < 0.001, figure 1(a)). The R 2 of the piecewise regression model is 0.75 with a wind speed decreasing trend of −0.060 m s −1 decade −1 during 1990-2012 (P < 0.001, inset of figure 1(a)). Following the reversal, the increase in wind speed has a fourfold higher magnitude during the rest of the decade (0.223 m s −1 decade −1 ; P < 0.001, inset of figure 1(a)). Specifically, 59% of the stations show increasing trends and 41% of them are significant (P < 0.05; figure S7). To ensure that the turning point is not caused by strong variations in wind speed at individual stations, we also performed piecewise linear regression on 300 data subsets, each containing 268 stations (40% of the total) that were selected randomly from the whole dataset (grey lines in figure 1(a)). All the subsamples show significant turning points with P < 0.001; and R 2 exceeds 0.55 for 96% of the linear regression models. The 95% confidence interval for the turning point location is also 2012 for the subsamples.
To test the robustness of the reversal of wind speed trend in China, we repeated our analysis using other station-based and quality-controlled datasets, including the Global Surface Summary of the Day from the U.S. National Oceanic and Atmospheric Administration (GSOD), the Hadley Integrated Surface Database from the U.K. Met Office Hadley Centre (HadISD), and a private dataset from the Geographic Data Sharing Infrastructure of Peking University (GDSI). All the datasets show a rapid increase in the 2010s ( figure 1(b)), which again is the period when installed wind capacity in China became prominent [1,4,5].

Spatiotemporal variability of the wind speed recovery
The wind speed recovery varies significantly for different regions of the country ( figure 2(a) 2 and S8) indicates that a reliable estimation of wind power gain requires fine-scale analysis both spatially and temporally (e.g. by hourly and for smaller area).

Relationship between capacity factor increase and wind speed recovery
We found that the overall CF in China (CF real ) is highly correlated with the cube of wind speed (R = 0.86, P < 0.001; figure 3), demonstrating the non-negligible role of wind speed in the interannual variability of CF for wind generation. Especially, during 2012-2019, the cube of wind speed in China increased significantly by 26% (R 2 = 0.86, P < 0.001), and the CF increased by 56% (R 2 = 0.83, P = 0.002), indicating that the recent recovery of wind speed is likely improving the efficiency of installed wind turbines. For installed wind turbines with fixed rated power, increasing wind speeds generate more wind power, increasing the CF.

Estimation of capacity factor increase boosted by wind speed recovery
To estimate the wind power gain and the CF growth solely driven by the increase in average wind speed, we calculated the CF raw with all other variables controlled (e.g. turbine technology improvement, wind farm expansion; see detail in Methods). We found that CF raw increased by 17% (from 0.1088 in 2012 to 0.1260 in 2019), accounting for 22.0% of the CF real increase (from 0.1468 in 2012 to 0.2249 in 2019). After calibration, CF cali increase equals 39.3% of the CF real increase. The corresponding increase in electricity generation is 31.65-56.49 TWh yr −1 (the lower and upper boundary are based on CF raw and CF cali accordingly; the following is the same) based on the total installed capacity in 2019 (210.05 GW, table 1). Both the upper and lower quartile in figure 4, corresponding to the provinces with abundant and limited wind resources respectively, experienced a slightly decreasing trend followed by a recovery. However, this assessment is not robust because the time series are short. At the province level, although the records are available only since 2013, we found that the trend of our calculated CF is consistent with values reported in most provinces (figures 5, 6 and S9). Considering that our calculation only uses the wind speed data and a single turbine type, the consistency provided additional convincing evidence that wind speed is a key factor attributing to the change in CF. Furthermore,  we also tried other wind turbine types (table S2) and found that the trend changes were robust regardless of which turbine model was selected (figure 7). With increasing installed capacity, the gain of wind power brought by the recovery of wind speed is potentially large. We set three scenarios to quantify the wind energy production potential. In the first scenario, which holds wind speed at the current level (2.20 m s −1 in 2019), the wind energy production will be 3311.3-5910.3 TWh yr −1 in 2060, equal to 40%-71% of the electricity consumption of China in 2021 (8312.8 TWh yr −1 [35]). The second scenario considered that wind speed might be influenced by decadal periodicity in ocean-atmosphere circulation [36] and potentially return to its historically highest level in 1969 (2.71 m s −1 , figure S10). For this hypothetical scenario, the wind power production would be 4501.8-8035.2 TWh yr −1 in 2060, equaling the production of 40-72 Three Gorges dams (111.8 TWh in 2019 [35]). The third scenario considered wind speed would decrease as IPCC predicted [37]. We took the historically lowest level (2.04 m s −1 ) in 2012 as the potential wind speed. In this case, the 3000 GW capacity will only generate total electricity of 2859.3-5103.5 TWh yr −1 (figure 7, table 1).

Discussion
All four data series mentioned above show upward trends in the 2010s ( figure 1(b)). However, the turning points visible in the CSD and GDSI datasets from CMDC occur around 2010, which is about one decade later than those seen in the GSOD and HadISD datasets (around 2000). This dissimilarity may stem from differences in the number and distribution of stations (table S1, figure S1). However, we also found that some stations with the same station ID in the HadISD and CSD datasets have time series abruptly become dissimilar following a long-term consistency. This discrepancy may be caused by: (a) stations being merged in HadISD [38] and (b) station relocation which has been processed with a spatial consistency check and corrected by CSD [27], but not by HadISD. Through the comparison of some overlapping stations in HadISD and CSD, the CSD data tend to have fewer such abrupt changes (figure S11).
The reason for the wind recovery remains unclear, leading to uncertainty about the future wind speed trend and wind power potential [37]. The driver of long-term wind speed decrease in China has two facets: (a) reduction of driving force associated with the change of large-scale atmospheric circulation [9,25,39] and (b) increase in resistance associated with land use/land cover change [13,21,40,41]. Considering that the land change trend is unlikely to reverse suddenly in 2012, the increasing wind speed is potentially a result of changed atmospheric circulation change [9]. Ongoing research is exploring these relationships and the associated physical processes. For example, the West Pacific Index (WPI) shows a more positive pattern during the period of wind speed decreases and a more negative pattern when wind speed recovers [42]. Negative phases of WPI mean more Western North Pacific tropical cyclones move into southern China and bring wind [43]. The increasing negative phases of Southern Oscillation Index (SOI) during 1975-2000 means colder sea water, smaller land-ocean air pressure gradient and weaker winter winds in China [16,44]. While the recovery of SOI positive phases since 2000 is followed by the reversal of wind speed. However, the relationship between the East Asian winter monsoon and SOI is unstable [45]. By now, no model fully reproduces the wind speed historical trends [37]. Additional studies in this field are needed considering the rapid growth in installed wind power globally.
Our analyses are based on a singular 2.5 MW wind turbine model. To address the uncertainty, we also calculated the CFs of turbines with rated capacities of 1.5 MW, 3.2 MW and 7.6 MW. The result shows that the 3.2 MW and 2.5 MW wind turbine models have much higher CFs than the 1.5 MW and 7.6 MW models for all the three wind speed scenarios (figure 7). Considering that 3.2 MW turbines can produce more electricity than 2.5 MW turbines for the same CF, this model is arguably an ideal choice for the observed/predicted wind trends instead of larger capacity of a 7.6 MW turbine. Choosing appropriately sized turbines based on wind situation can help narrow the CF gap between China and the United States [8,9,46].

Conclusion
Our analysis shows that following decades of decreasing wind speeds, wind speeds have recovered in most places in China in the 2010s. This reversal in wind speed trend has significant implications for wind power production. We used an innovative method to calculate the wind speed-driven CF change and found that this recovery has increased the CF for wind generation in China by 17%, explaining almost 22.0%-39.3% of the reported increases in the CF over 2012-2019. This change in wind speed allows wind farms to be more efficient sources of electricity generation, and therefore, supports the role of wind energy development to help replace fossil fuels in the future. However, the underlying cause of wind speed change remains unclear, lending uncertainty to the prediction of future wind speeds. To explore this uncertainty, we used three possible scenarios related to wind speed changes (reduction, increase and stay). But further research is still required to make a more informed conclusion. An improved understanding of the drivers of changes in wind speed can provide a basis for improved regulation of wind farms siting, turbine selection and management.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors.