Attribution of Terrestrial Near‐Surface Wind Speed Changes Across China at a Centennial Scale

Near‐surface wind speed (NSWS) over China shows multiple time‐scale changes at a centennial scale, but the contributions of internal variability (IV), anthropogenic forcing (ANT), and natural forcing (NAT) to those changes remain unknown. This study investigated the contributions of IV, ANT, and NAT to NSWS changes at a centennial scale. Results show that the NSWS changes were attributed mainly to IV. IV not only modulated the interannual changes in NSWS but also determined the interdecadal transition in NSWS. The relative contributions of IV to the interannual and decadal NSWS exceeded 75.0%. ANT contributed particularly to the long‐term reduction in NSWS; especially, it has contributed 55.0% of the reduction in NSWS since 1957, serving as the major contributor to the reduction in NSWS. NAT had a small‐to‐negligible effect on China's NSWS throughout the study period. This study enhances our understanding of NSWS changes at different time scales.


Introduction
Near-surface wind speed (NSWS) on land plays a pivotal role in atmospheric circulation and significantly influences the distribution and transportation of substances and energy in the lower atmosphere.Changes in NSWS have profound impacts across multiple environmental and climatic domains.These include dust storms (R. Wang et al., 2017), regional evapotranspiration (McMahon et al., 2013), hydrological conditions (McVicar et al., 2012), air quality (Y.Zhang et al., 2020), and wind power generation (Pryor & Barthelmie, 2021).Consequently, it is important to investigate the variability in NSWS.
In recent decades, most of the existing studies on NSWS have been into the stilling and the causes (Vautard et al., 2010;J. Wu et al., 2018;G. Zhang et al., 2020), the reversal of stilling (Zeng et al., 2019;Z. Zhang & Wang, 2020), and the interannual and interdecadal variations in NSWS (Chuan et al., 2024;Zha et al., 2022a).The NSWS changes could be influenced by internal variability (IV), anthropogenic forcing (ANT), and natural forcing (NAT).The ANT induced by human activities mainly includes land use and cover change (LUCC), greenhouse gas emissions, and aerosol emissions (Bichet et al., 2012;Chen et al., 2020).Hasselmann (1976) has suggested that different forcing signals leave particular fingerprints in the climate system, and by isolating these fingerprints, the main forcing factors influencing climate change could be determined.On this theoretical basis, detection and attribution (D&A) for climate change has been extensively performed (Zhai et al., 2018).Researchers have mainly emphasized the temperature (Najafi et al., 2015), rainfall (Delworth & Zeng, 2014), extreme weather and climate events (Herring et al., 2018), and atmospheric circulation (Kim et al., 2017), and influencing mechanisms and quantitative contributions of the IV and ANT on above-mentioned variables are revealed in depth (Fischer & Knutti, 2015).Nevertheless, as a key indicator of climate change, NSWS has rarely been subjected to attribution.
The optimal fingerprint is the most commonly used method to isolate the effects of IV and ANT on climate change (Hasselmann, 1997), which can be used to quantify the influence of a forcing factor on climate change based on a linear regression hypothesis (Ribes et al., 2017).However, NSWS variations are characterized by non-stationarity and strong fluctuations, meaning that using the optimal fingerprint for attribution analysis of NSWS changes could cause a large uncertainty.Furthermore, the optimal fingerprint is reliant on observed data; however, the lack of meteorological stations and the quality problems of observation records before 1970 makes it difficult to conduct an attribution analysis for NSWS at a long timescale based on the optimal fingerprint.The Detection and Attribution Model Intercomparison Project (DAMIP) is also utilized to investigate the impacts of different forcing factors on NSWS changes (Shen, Zha, Zhao, et al., 2021;Wohland et al., 2021).Nevertheless, as is well known, most global climate models (GCMs) show large biases when simulating the NSWS (J.Wu et al., 2020).Although GCMs may have acceptable biases at the global scale, it is difficult to apply them in regions with complex topography because of their coarse resolutions and being defective in simulating regional and local climate characteristics (Goyal et al., 2021).Moreover, the changes in NSWS are sensitive to the type of underlying surface types and the regional climate features (Zha et al., 2019).Therefore, the attribution of NSWS changes at the regional scale based on GCMs will generate large uncertainties (Gutowski et al., 2020;Luo et al., 2020).As an alternative and more physically based approach, some studies also use the dynamical downscaling technique with a regional climate model (RCM) to isolate the effects of main forcing factors on climate change (Huang et al., 2023).
The Sixth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC AR6) pointed out that the confidence about the causes of changes in NSWS is lower than those about changes in temperature and precipitation (IPCC, 2021).In a previous study, we found that the NSWS in China showed decadal characteristics at the centennial timescale (Shen, Zha, Wu, & Zhao, 2021).However, scientific issues remain to be clarified, such as the extent to which the interannual, interdecadal, and long-term changes in NSWS can be attributed to IV, ANT, or NAT.Consequently, there is an urgent need for attribution analysis of NSWS changes, which can deepen the understanding of the causes of NSWS changes and lead to a better comprehension of how human activities impact those changes.With that in mind, we investigated the effects of IV, ANT, and NAT on NSWS changes across China at the centennial timescale with the combination of the GCMs and the RCM.

Data Sets
A gridded daily mean observed data set over China (CN05.1) is employed to estimate the performance of an RCM in simulating the observed NSWS.CN05.1 has a spatial resolution of 0.25°× 0.25°and covers the period from 1961 to 2022, and it has been used widely for analyzing climate changes in China and estimating the performance of GCMs and RCMs (Bucchignani et al., 2016;J. Wu et al., 2012).We can find more details about how CN05.1 is created from J. Wu et al. (2012).For the lateral boundary conditions and initial fields of an RCM, the National Oceanic and Atmospheric Administration Twentieth Century Reanalysis Project Version 2c (NOAA-20CRv2c) is used (Compo et al., 2011).NOAA-20CRv2c is a comprehensive global atmospheric circulation data set covering the period from 1900 to 2014, assimilating the Hadley Center's observed surface pressure, sea surface temperature, and sea ice with a temporal resolution of 6 hr and a spatial resolution of 2°× 2°.NOAA-20CRv2c has good homogeneity and stationarity in the time series.Previous studies have shown that NOAA-20CRv2c is a high-quality data set for the climate research community for both model validation and diagnostic study (Compo et al., 2011;Slivinski et al., 2019).We also use the HistoricalMisc and HistoricalNat experimental outputs from DAMIP of CMIP5, because the DAMIP of CMIP6 has not been fully released when the present numerical simulations are designed and carried out.HistoricalMisc involves a set of anthropogenic forcings, including aerosols, greenhouse gases, land use, and ozone, while HistoricalNat involves a natural-only historical simulation.Table S1 in Supporting Information S1 shows the details of the six selected DAMIP models, which were chosen Geophysical Research Letters 10.1029/2024GL108241 for the following three reasons: (a) they provide all the variables needed to drive an RCM; (b) they have a relatively high horizontal resolution; (c) they contain at least three member runs.

Methods
The RCM used in this study is the Weather Research and Forecasting (WRF) model version 3.8.1.The simulation domain covers most of East Asia (0°-55°N, 57.5°-142.5°E)with a spatial resolution of 0.5°× 0.5°; see Zha et al. (2022b) for the simulated region.The vertical dimension is split into 33 levels, with its top fixed at 50 hPa.WRF has been integrated for 106 years from 1900 to 2005, and the first year of the simulation period ( 1900) is treated as a spin-up period.The parameterization schemes for the dynamical downscaling simulation with WRF (DDS-WRF) are the Thompson aerosol-aware microphysics scheme, the RRTMG long-wave and shortwave radiation schemes, the MYJ planetary boundary layer scheme, the BMJ cumulus parameterization scheme, and the Noah-MP land surface model.These physical parameterizations are considered appropriate for simulating climate change in East Asia (Y.Wang et al., 2020).Since the volcanic aerosols, ozone, and LUCC have a considerable effect on long-term climate change (Cooper et al., 2018;Diallo et al., 2017), the WRF model is modified to include both spatially and temporally varying external forcing (Text S1 in Supporting Information S1).Three dynamical downscaling experiments with different lateral boundary conditions and external forcing signals are designed to isolate the effects of IV, ANT, and NAT on changes in NSWS at the centennial scale: (a) the all-forcing experiment (EX ALL ), (b) the ANT experiment (EX ANT ), and (c) the NAT experiment (EX NAT ) (Text S2 in Supporting Information S1).
Two independent variables can be strongly correlated but not necessarily share a cause-and-effect relationship (Hagan et al., 2019;Liang, 2016), so the Liang-Kleeman information flow is used to examine the causality between two variables and validate its robustness (Text S3 in Supporting Information S1).Johnson's relative weight analysis (RWA) method is used to quantify the relative contributions of IV, ANT, and NAT to the NSWS changes (Johnson, 2000) (Text S4 in Supporting Information S1), which has been extensively used to estimate the contribution of IV, ANT, and NAT to the climate change (Huang et al., 2023;R. Li et al., 2021).A piecewise linear function (PWLF) is used to identify the transition point of NSWS at the centennial scale (Jekel & Venter, 2019).Details about the PWLF method are shown in Text S5 in Supporting Information S1.We also use the ensemble empirical mode decomposition (EEMD) to extract interannual and decadal signals of NSWS (Z.H. Wu & Huang, 2009).Based on EEMD, the NSWS (Y(t)) is separated into three components (Y(t) = H(t) + A (t) + L(t)).H(t) is a quasi-annual period (interannual signal), A(t) is a low-frequency component (interdecadal signal), and L(t) is a long-term trend.To quantify the consistency of temporal evolutions between two variables, we calculate the probability of extremes appearing at the same time point (PEST) between two variables (Zha et al., 2019).To evaluate the performance of the DDS-WRF in simulating the NSWS changes, we calculate the skill score, root-mean-square error (RMSE), and correlation coefficient.

Performance of DDS-WRF in Simulating the NSWS Changes Across China
The performance of DDS-WRF in simulating NSWS changes across China is estimated first.Both NOAA-20CRv2c and DDS-WRF capture the main characteristics of NSWS climatology, that is, high values of NSWS in the Tibetan Plateau and north China and low values of NSWS in southeast and southwest China (Figure S1 in Supporting Information S1).However, the NOAA-20CRv2c overestimates the NSWS over southeast China and underestimates it over the Tibetan Plateau, and DDS-WRF consistently overestimates the NSWS over China in its entirety.Observation and DDS-WRF (NOAA-20CRv2c) are mainly correlated positively, and the correlation coefficients and skill scores show a consistent spatial distribution; however, the correlation coefficient and skill score between observation and DDS-WRF are higher than those between observation and NOAA-20CRv2c (Figure S2 in Supporting Information S1).The correlation coefficients (skill scores) between observation and NOAA-20CRv2c, and between observation and DDS-WRF are 0.21 (0.27) and 0.35 (0.40), respectively.
The observed NSWS shows a downward trend from 1970 to 2005 (Figure 1a), which is similar to the findings of previous studies (Ge et al., 2021;X. Li et al., 2022).The reduction in NSWS simulated by NOAA-20CRv2c (Figure 1b) is weaker than those simulated by DDS-WRF (Figure 1c).The fact that the NSWS reduction simulated by DDS-WRF is stronger than those simulated by NOAA-20CRv2c could be due to considering the LUCC effects in WRF's Noah-MP land surface model, as the surface roughness change induced by LUCC has a pronounced effect on the long-term changes in NSWS (Bichet et al., 2012;Vautard et al., 2010).When the longterm trend of NSWS is removed, the detrended NSWS simulated by DDS-WRF is closer to observation than those in NOAA-20CRv2c (Figure 1d).If we just consider interannual and interdecadal changes of the detrended NSWS, a stronger correlation, and a smaller RMSE are presented in the DDS-WRF compared to NOAA-20CRv2c.For interannual NSWS, the correlation coefficient and RMSE between DDS-WRF (NOAA-20CRv2c) and observation are 0.74 (0.67) and 0.02 m s 1 (0.04 m s 1 ), respectively (Figure 1e).For interdecadal NSWS, the correlation coefficient and RMSE between DDS-WRF (NOAA-20CRv2c) and observation are 0.90 (0.21) and 0.04 m s 1 (0.06 m s 1 ), respectively (Figure 1f).Particularly, DDS-WRF can capture the turning points of NSWS in 1978 and 1992, whereas NOAA-20CRv2c does not.These results imply that DDS-WRF can well reproduce the interannual and interdecadal changes of the observed NSWS, although the NSWS's long-term trend is somewhat underestimated.

Interannual and Interdecadal Changes in NSWS Are Caused by IV
Temporal evolutions of NSWS with different forcings are shown in Figure 2. The NSWS shows an upward trend from 1900 to 1956 (0.018 m s 1 decade 1 , p < 0.01) and a downward trend from 1957 to 2005 ( 0.011 m s 1 decade 1 , p < 0.01) (Figure 2a), and this feature is also found in other reanalysis products (e.g., ERA-20C, ERA-20CM, and CERA-20C) (Shen, Zha, Wu, & Zhao, 2021).Temporal changes in NSWS with IV are similar to those with ALL (Figure 2b), suggesting the IV is responsible for the NSWS changes at the centennial timescale.No significant decadal transition in NSWS is found with ANT and NAT.With ANT, NSWS mainly displays a decrease ( 0.01 m s 1 decade 1 ; p < 0.01) (Figure 2c), suggesting ANT mainly induces the long-term reduction in NSWS.With NAT, neither decadal transition nor long-term variation in NSWS occurred (Figure 2d), suggesting NAT determines neither the decadal changes nor long-term trends of NSWS.Spatially, the variability in NSWS with IV is similar to those with ALL (Figures S3a and S3b in Supporting Information S1), whereas the ANT and the NAT significantly underestimate interannual and interdecadal fluctuations in NSWS (Figures S3c and S3d in Supporting Information S1).The correlation coefficient and skill score of NSWS between ALL and IV are close to unity, and the PEST between ALL and IV reaches 90.0% (Figure S4 in Supporting Information S1).Furthermore, the probability density function of NSWS with IV is also closer to those with ALL compared to ANT and NAT (Figure S5a in Supporting Information S1), and NSWSs with IV and ALL show a significant causality, as the information flow from IV to ALL is larger than those from ANT and NAT to ALL (Figures S5b and S6 in Supporting Information S1).All these results confirm that IV is a predominant factor inducing the NSWS changes over China at the centennial timescale.
Because IV not only determines the decadal transition in NSWS but also influences the amplitude of fluctuation in NSWS, we calculate the variance contributions of different forcing factors to NSWS changes on interannual and interdecadal scales.Interannual and interdecadal variance contributions of IV to total NSWS changes exceed 80.0% (Figure 3a).By contrast, the effects of ANT and NAT on interannual and interdecadal NSWS are weak (<20.0%).Particularly, NAT has a small-to-negligible impact on the interannual and interdecadal changes in NSWS (<10.0%).We also carry out a RWA to quantify the contributions of different forcings to the NSWS changes (Figure 3b).The IV can explain the majority of the interannual and interdecadal variations in NSWS (>70.0%),especially for interannual NSWS, the IV's contribution exceeds 82.0%.The relative contributions of ANT and NAT to interannual and interdecadal changes in NSWS are less than 20.0%.Given that NSWS has a considerable transition, we compare the relative contributions of different forcings to NSWS changes during two different periods (the turning point is validated based on the PWLF method) (Figure 3c).The results show that IV has been a major contributor to interannual and interdecadal changes in NSWS from 1900NSWS from to 1956NSWS from and 1957NSWS from -2005.However, since 1957, the IV's contribution declined, and the ANT's contribution enhanced.These results imply that the effects of ANT on interdecadal changes in NSWS could have strengthened since 1957.Therefore, to improve the accuracy of the model in predicting NSWS changes, we should focus on not only the performance of the model in simulating the IV but also its sensitivity to ANT effects.

ANT has a Significant Effect on the Long-Term Slowdown in NSWS
Long-term trends of NSWS with different forcing factors from 1900 to 2005 are investigated.NSWSs with ALL (0.004 m s 1 decade 1 ; p < 0.10) and IV (0.011 m s 1 decade 1 ; p < 0.10) show the upward trend (Figure S7a in Supporting Information S1).It is worth noting that the NSWS with ANT shows a significant downward trend, reaching 0.007 m s 1 decade 1 (p < 0.10).The trend of NSWS with NAT is weak and not significant.These results highlight the importance of ANT in NSWS reduction.Similarly, from 1900 to 2005, the spatial pattern of NSWS trend with ALL (Figure S7b in Supporting Information S1) is the same as those with IV (Figure S7c in Supporting Information S1).Meanwhile, the NSWS with ANT mainly shows a downward trend across the whole of China (Figure S7d in Supporting Information S1), especially in Northeast China and the middle and low reaches of the Yangtze River; whereas, a significant long-term trend is not found with NAT (Figure S7e in Supporting Information S1).
Because NSWS shows a significant transition at a centennial scale and ANT has a persistent slowing effect on NSWS, we further clarify the contributions of different forcing factors to the long-term trends of NSWS.The effects of ANT on the interannual and interdecadal change in NSWS are not significant, but ANT becomes more important in determining the long-term trend of NSWS.Before 1957, the upward trend of NSWS as shown in Figure 3d is dominated mainly by IV, and the contributions of ANT and NAT to the long-term trend of NSWS are smaller (<20.0%).After 1957, ANT accounts for approximately 55.0% of the trend of NSWS, and the contributions of ANT to the trend of NSWS are larger than IV.During two different periods, NAT has always had a small contribution to the trends of NSWS (<10.0%).Spatial patterns of NSWS trends with ALL are the same as those with IV from 1900to 1956and 1957to 2005 (Figure 4).The NSWS experiences an increasing trend from 1900 to 1956 over China except for the Tibetan Plateau, which experiences a decreasing trend from 1957 to 2005.Compared to the NSWS trends with IV, the increase in NSWS from 1900 to 1956 and the reduction in NSWS from 1957 to 2005 are due mainly to IV (Figures 4c and 4d).ANT mainly induces the slowdown in NSWS over most regions of China from 1900 to 2005, especially for the Tibetan Plateau, central China, northern China, and the Shandong Peninsula (Figures 4e and 4f).Compared to the NSWS trends with ALL, ANT has been a contributing factor to the weakening of NSWS over the Tibetan Plateau, northeast China, north China, and the middle and low reaches of the Yangtze River from 1900 to 1956.NAT had no decisive effect on long-term changes in NSWS from 1900 to 2005 (Figures 4g and 4h).

Conclusions and Discussion
In this study, we revealed the contributions of IV, ANT, and NAT to the NSWS changes across China at the centennial timescale based on a joint use of the GCMs and the RCM.Compared to the outputs of the reanalysis product, DDS-WRF shows better performance at capturing the spatiotemporal features of observed NSWS in China, especially for the detrended NSWS, interannual, and interdecadal variations in NSWS.NSWS shows a significant decadal transition at the centennial scale, increasing from 1900 to 1956 and decreasing from 1957 to 2005.The NSWS changes across China at the centennial scale are attributed mainly to the IV, which not only determines the interannual changes in NSWS but also controls the interdecadal transition in NSWS.The variance contributions of IV to the interannual and interdecadal NSWS at the centennial scale exceed 80.0% from 1900 to 2005, and the relative contributions of IV to the interannual and interdecadal NSWS are more than 70.0%.ANT mainly causes the NSWS reduction, especially for the long-term slowdown in NSWS since 1957, whose relative contribution could reach 55.0%.Additionally, NAT has mall-to-negligible impacts on the NSWS changes in China during the whole study period.
We also compare the NSWS changes with some large-scale circulation indices to discover the circulation factors in the IV that may induce the interdecadal changes in NSWS (Table S2 in Supporting Information S1).The selected large-scale circulation factors and NSWS must satisfy causality.Finally, AO and NAO are selected.AO (NAO) and NSWS show inversely phased changes with a correlation coefficient of 0.70 ( 0.81), and both AO (NAO) and NSWS show a similar decadal transition.The information flow from AO (NAO) to NSWS is significant, and the uncertainty range of information flow is small (Figure S8 in Supporting Information S1).These results indicate that in the IV of the climate system, AO and NAO may be the factors inducing the decadal changes in NSWS across China.However, the physical mechanisms of these circulation systems influencing China's NSWS changes need to be further investigated, and it is beyond the scope of this study.Also, recent studies suggest that the IV (e.g., AO) could be partly affected by external forcing factors (Hua et al., 2018;Qin et al., 2020).Therefore, it also remains a challenge to quantify the long-term contribution of external forcing factors to NSWS changes on the regional scale by modulating the predominant mode of IV.
In this study, the NSWS trend simulated by DDS-WRF (NOAA-20CRv2c) accounts for 45.7% (17.4%) of the trend of observed NSWS.Therefore, DDS-WRF better captures the long-term trend of the observed NSWS than the reanalysis product, which could be because the DDS-WRF is modified to consider spatially and temporally varying forcing signals (e.g., aerosols, ozone, solar constant, and LUCC).Nevertheless, the underestimation of the decreasing trend in NSWS remains to be solved.Consequently, the relative contributions of IV and ANT to the NSWS trend could be underestimated.The improved simulation performance with a DDS-WRF indicates that we must consider the observed and temporally varying external forcings and carry out high-resolution numerical simulations.Furthermore, the observed NSWS data sets should also be considered for assimilation in an RCM.These works can further improve the ability of the model to simulate the NSWS trend, and therefore the uncertainty of the contributions of different forcings to the NSWS trend can be quantified.Additionally, the long simulation period and limited computing resources hinder high-resolution nesting simulations in some special areas, but it is important to quantify the regional differences of IV and ANT impacts on NSWS changes.In future work, nested high-resolution numerical simulations will be needed for some typical regions of China, facilitating a more objective and accurate evaluation of the regional differences for those forcing factors affecting the spatiotemporal changes in NSWS.

Figure 1 .
Figure 1.Temporal changes of normalized near-surface wind speed (NSWS) in observation (a), NOAA-20CRv2C (b), and DDS-WRF (c) over China from 1970 to 2005.(d) Is the same as (a), but for the detrended NSWS sequences.(e, f) Is the same as (d), but for the interannual sequence and decadal sequence of NSWS, respectively.In (a)-(c), the red lines denote the linear fitting, and the shaded regions represent the ranges of the 25th and 75th percentiles of the normalized NSWS in all grids.In (d)-(f), r is the correlation coefficient between DDS-WRF (NOAA-20CRv2C) and observation, and p < 0.05 (>0.1) denotes the r passes (fails) a significance t-test at the 0.05 (0.10) level.The interannual and decadal signals are extracted based on the ensemble empirical mode decomposition method.

Figure 2 .
Figure 2. Temporal changes of area-weighted mean near-surface wind speed (NSWS) anomaly from 1900 to 2005 with ALL (a), IV (b), ANT (c), and NAT (d) (unit: m s 1 ).Shading represents the ranges of the 25th and 75th percentiles of NSWS in all grids over China.In (a) and (b), blue and red lines denote the piecewise linear function (PWLF) fitting curves.In the insets, the red and blue bars denote the trends of NSWS from 1900 to 1956 and 1957 to 2005, respectively, and p < 0.05 denotes the PWLF fitting passes the 0.05 significance level.Pink dotted lines denote the decadal changes in NSWS, which are obtained from 31-year time series low-pass filtered with a Gaussian-type filter.

Figure 3 .
Figure 3. Contributions of different forcing factors to near-surface wind speed (NSWS) changes.(a) The variance contributions of IV, ANT, and NAT to interannual and interdecadal changes in NSWS from 1900 to 2005.(b) Is the same as (a), but for relative contributions.In (b), bootstrap sampling with replacement is performed 300 times on the spatial fields of the combinations of intrinsic mode functions (IMFs) of ensemble empirical mode decomposition to evaluate the uncertainty range of the relative weights of IV, ANT, and NAT.(c) Is the same as (b), but for two periods 1900-1956 and 1957-2005.(d) The trend contributions of different forcing factors on long-term changes in NSWS during 1900-1956 and 1957-2005.In (a) and (d), the black bar denotes the 95% confidence intervals of the results.In (b), the median (the center line), the mean value (dots), the upper (75th) and lower (25th) percentiles (box limits), and the 1.5 times the interquartile range (whiskers) are presented.