Simulated Influence of Mountainous Wind Farms Operations on Local Climate

Due to the complex terrain in mountainous areas, the local climate is also more complex. To be specific, the increase of altitude is conducive to the convergence of horizontal wind and the development of vertical upward movement on the windward slope (Liao et al. ^[1]). The change of near surface wind speed is closely related to the terrain, and the mountain terrain wave is invariant to the terrain (Hu et al. ^[2]). Terrain-induced gravity waves are able to affect the strength of convective activity, and the propagation of convective system may also be modulated by gravity waves (Huang et al. ^[3]). Topography can significantly change the boundary layer flow, affect the local weather and circulation, form valley wind, and strengthen the inversion, foehn wind, canyon wind and topographic cloud in the valley at night (An et al. ^[4]; Miao et al. ^[5]). The imbalance of solar radiation over mountains is related to the convective core over mountains (Tian et al. ^[6]).

According to the Outline of the 14th Five-Year Plan for the National Economic and Social Development and the Long-Range Objectives Through the Year 2035, the time during the implementation of the plan (2021-2025) is the main window period for China to reach peak carbon emissions and during this period China will focus on building a new power system with clean energy. The capacity of the installed wind power in China reached 209.94 GW by the end of 2019, and the number is bound to increase in the future. The operation of a wind turbine (WT) increases the surface roughness, extracts kinetic energy from the atmosphere, reduces the wind speed, and changes the wind shear and the spatial distribution of pollutants (Chen et al. ^[7]; Mo et al. ^[8]). At the same time, WTs can indirectly affect the vertical profile, the exchange of material and energy between the surface and atmosphere, the stability of the boundary layer, and other factors such as temperature and precipitation (Zhou et al. ^[9]; Jiang et al. ^[10]; Li et al. ^[11]; Abbasi et al. ^[12]). Therefore, large-scale development of wind farms may cause a wide range of changes in weather and climate (Roy et al. ^[13]).

Previous research on the effect of wind farms on climate has been conducted by using observational data and satellite remote sensing data (Smith et al. ^[14]; Armstrong et al. ^[15]; Zhou et al. ^[16]; Harris et al. ^[17]; Liu et al. ^[18]; Yue et al. ^[19]). Numerical simulation is an important means of evaluating the effects of wind farms on climate. At present, two main factors are considered for the numerical simulations of wind farms: (1) the increase in surface roughness length (Keith et al. ^[20]; Kirk-Davidoff et al. ^[21]) and (2) momentum sinks and increase in turbulent kinetic energy (TKE) sources (Fitch et al. ^[22]; Fitch et al. ^[23]). Fitch ^[24] pointed out that the amplitude of variation near surface meteorological variables, which was simulated by increasing the surface roughness, deviated greatly from the observations. Therefore, the momentum sink and the increase in TKE sources are currently used for simulations. Fitch et al. ^[22] developed a WT module in the Weather Research and Forecasting model (WRF), which is known as the WRF-Fitch, and it has become the main simulation tool for studying the effect of wind farms on climate. Xia et al. ^[25] used the WRF-Fitch model to simulate LST before and after the establishment of large wind farms in central and western Texas and analyzed the vertical changes in temperature due to these wind farms. They further discussed the average regional vertical profile of temperature changes caused by wind farms. Wang et al. ^[26] used the WRF-Fitch model to further explore the impact of wind farms on wind speed and wake effect in different atmospheric stabilities. The results showed that in an unstable atmospheric structure, the wake effect quickly mixes and dissipates downstream, causing the downstream wind speed to recover as soon as possible. However, in a stable atmospheric stratification, the wake effect significantly weakened the wind speed. Mangara et al. ^[27] simulated the impact of wind farms along the coast of the Bohai Sea on climate using the WRF model. The results showed that the wind speed decreased by more than 4% within 10km downwind of the WT. At a height of 150 m or 25 km downwind, the wind speed decreased by ~2%, which can be ignored. Li et al. ^[28] simulated the impact of large wind farms on temperature in northern China and found temperature may cause the attenuation of wind speed in the wind farms and adjacent areas, with the attenuation significantly stronger in winter (January) than that in summer (July). Sun et al. ^[29] used the WRF-Fitch model to separate the effects of the momentum sink and the TKE source. The momentum sink is the main factor causing wind speed attenuation in the wind farms, resulting in less vertical wind shear, and the TKE source does the contrary. The impact of wind farms on wind speed and TKE at the hub height is usually local.

The above studies constitute good basis for further research on the effects of wind farm on local climate. However, to the best of our knowledge, such studies in China have only focused on the large-scale wind farms in the plains in northern China, whereas little work has been done on the wind farms distributed in the complex mountainous areas in southern China, although the number of these wind farms has gradually increased with the rapid development of wind energy industry. Therefore, in the present research, the Dawu and Suizhou wind farms in the north of Hubei Province were selected as the subjects for simulation, and the area (113° 07′ 30″E-114° 37′ 16″E, 31° 09′ 19″N-32° 25′ 04″N) was chosen as the representative of the mountainous regions in southern China. With such efforts, the present study is expected to provide scientific support for the future utilization of wind energy resources and facilitate the plan of the layout of climate-friendly wind farms in the complex terrain in southern China.

The meteorological background field was obtained from the FNL reanalysis data from the National Centers for Environmental Prediction (NCEP). The time period studied was from January 1 to January 31 (winter) and July 1 to July 31 (summer), 2014. The temporal resolution is 6h, and the spatial resolution is 1°×1°.

The observational data from the Anlu Meteorological Station in Hubei Province were used to test the simulations of the numerical model from the China Integrated Meteorological Information Service System developed by the National Meteorological Information Center. The data include hourly temperature, wind speed, and relative humidity for the abovementioned period of study. There are 744 hourly data samples in January and July, respectively. The hourly observational data and numerical simulation data are converted to daily data (31 samples every month). The correlation significance test and root mean square error of the two sets of data were calculated.

The correlation coefficient (R) is a statistical indicator used to reflect the closeness of the correlation between the two sets of variables. Mean bias (MB) can show average deviation between two groups of values. The R value and MB between the observed value of the Anlu Meteorological Station and the simulated value were used to test the WRF model simulations (Sun et al. ^[29]; Ning ^[30]).

where x_s is the simulated value and x_o is the observed value.

The t value was used to test the degree of variation in meteorological parameters caused by the effect of wind farms. There are multiple sets of time series values at each simulation grid, corresponding to the control ${\left( {x_1^{{\rm{CT}}}, x_2^{{\rm{CT}}}, \cdots , x_n^{{\rm{CT}}}} \right)}$ and the sensitivity ${\left( {x_1^{{\rm{SCEN}}}, x_2^{{\rm{SCEN}}}, \cdots , x_n^{{\rm{SCEN}}}} \right)}$ experiments. The sequence x^d represents the difference between the sensitivity and the reference tests $\left( {x_i^d = x_i^{{\rm{SCEN}}} - x_i^{{\rm{CT}}}, i = 1, 2, \cdots , n} \right)$. The t-value is calculated as:

where n is the number of samples. The full sample contains 744 values (hourly data in one month), and 124 values during the day or night (day: 11:00-14:00, night: 22: 00-01: 00 (local time)); d represents the average value of the physical quantity in the time series, and S_D represents the standard deviation of the physical quantity on the time series. The t-value on each grid is calculated by using Equation (3) and compared with the value corresponding to t_0.1 to determine the grid with a 90% confidence.

The present work used the WRF(V3.8) model to study the impact of wind farms on local climate in the mountains. The model adopted three layers of two-way nesting, and the simulation center was located at 31.81° N, 113.90° E. The horizontal resolution of the d01 area was 15km, and the number of grids was 90 × 90. The horizontal resolution of the d02 area was 5km, and the number of grids was 97×97. The horizontal resolution of the d03 area was 1km, and the number of grids was 141× 141 (Fig. 1). The d03 area covered the Suizhou and Dawu wind farms in the northern part of Hubei Province. There were 43 layers in the vertical direction, including 23 layers below 1km and 8 layers within the sweeping range of the WT blades. During the period of study, the WRF model ran every 7 days using the FNL global reanalysis data as the initial and background fields, and each duration was continuously integrated for 7 days. The last day of the previous simulation overlapped with the first day of the next simulation. The first 24 h of each simulation result was used as the model spin-up time, and the simulation results of the next 6 days were retained. The entire month simulation was completed after five cycles.

Figure 1.  WRF simulated domain.

According to previous studies, the combination of the physical parameterization schemes (Table 1) can better reflect China' s near-surface climate characteristics (Wang et al. ^[26]; Sun et al. ^[29]).

Since the original model ignored the system loss and significantly overestimated the turbulence source, the Fitch module was updated in June 2020 and a turbulence correction factor was introduced to advect TKE generated by the WT. In this study, the updated Fitch module was used for numerical simulation, and the turbulence correction factor was set as 0.25 (Archer et al. ^[36]).

Figure 2 shows the variation in the 2m temperature, 2m relative humidity, 10m wind speed of the Anlu Meteorological Station (red line), and control experiments value (black line) for January and July 2014. It was observed that the simulation values were similar to the observed values. In January, the average deviations were -0.34 ℃, -4.45 %, and 0.93 m s^-1, and their correlation coefficients were 0.79, 0.81, and 0.78, respectively. In July, the average deviations were - 0.06 ℃, - 2.19 %, and 0.59 m s^-1, and their correlation coefficients were 0.82, 0.79, and 0.47, respectively. All the correlation coefficients passed the 99% confidence level, indicating that the WRF model shows good simulation for this area. Therefore, the WRF model can meet the requirements of this study and effectively simulate near-surface meteorological parameters.

Figure 2.  Observed (red line) and simulated (black line) 2m temperature (a, Jan; d, Jul), 2m relative humidity (b, Jan; e, Jul), and 10m wind speed (c, Jan; f, Jul) at Anlu station.

A sensitivity experiment was set up based on the actual WT installation in the Dabie Mountains. The study area was mountainous, with mountains extending along the northwest-southeast direction. The highest and lowest elevations in the region were 737 m and 37m, respectively, with height difference exceeding 700 m. The WTs were installed on the ridges. The WTs of the Suizhou wind farm (WF_sz) were located at a higher altitude (averaging 538 m) than Dawu wind farm did (WF_dw, averaging 391 m). The highest WF_sz was at 737 m, and the lowest at 218 m. The highest WF_dw was at 700 m, whereas the lowest was at 134m. The background wind field in the study area during winter was distributed divergently in the WF_sz, with a relatively high wind speed in the ridge area. There was a north-south airflow channel in the WF_dw, with a higher wind speed at night than that during the day. In summer, the prevailing wind was in the southeast direction, being higher in the ridge area, which was slightly less than that in winter. The wind speed at night was higher than that during the daytime.

The main type of WT (GW115/2000 kW) installed in this area was input into the WRF model, and the parameters of this WT were: hub height = 85 m, blade diameter = 115 m, and power = 2000 kW. The WT thrust coefficient and power curves are shown in Fig. 3. By the end of 2016, 18 wind farms have been built, with an installed capacity of approximately 870, 000 kW. A total of 507 WTs were studied, which were a part of the Tongbaishan-Dabieshan wind farm group. A total of 278 WTs were installed in the WF_sz, and 229 WTs were installed in the WF_dw. The distribution of the WTs is shown in Fig. 4.

Figure 3.  Thrust coefficient curve (red line) and power curve (blue line) of the turbine used in the numerical stimulation.

Figure 4.  Layout position of wind turbines (solid dots) and terrain height (color map, units: m) in d03 (left: Suizhou wind farm; right: Dawu wind farm).

At hub height, the wind speed (WS) at the hub height (85 m) had different degrees of attenuation in the WF in winter, with the attenuation being greater at night than in the day. The intensity and range of WS attenuation in WF_dw were weaker than those in WF_sz. In WF_sz, the WS variation during the daytime (Fig. 5a) was as high as 0.7 m s^-1 with 27 grids passing the 90% significance test, whereas the WS in WF_dw dropped by 0.2-0.4 m s^-1. The variation in WS was more obvious at night (Fig. 5b) and had an expanded range. The WS in the WF_sz dropped by up to 1.1 m s^-1, and 60 grids passed the 90% significance test. The grids that passed the test were located around the WTs. The area where the WS decreased was in and around 18 km area of the WF_sz. The WS decreased by up to 1.0 m s^-1 in WF_dw, but this variation did not pass the significance test.

Figure 5.  Difference in mean wind speed (m s^-1) at hub height (85 m) between sensitivity and control experiments in winter. (a) Daytime and (b) nighttime.

In summer, WS attenuation at the hub height is more obvious than that in winter, and it is more significant in the summer nights. In the WF_sz, the WS during summer daytime decreased by 0.2-0.8 m s^-1, and a total of 43 grids passed the 90% significance test (Fig. 6a), whereas at night it decreased by 0.2-1.1 m s^-1 (Fig. 6b), and a total of 199 grids passed the significance test. The grids that passed the test were mainly those with the WTs and the areas surrounding them. The area of attenuation was mainly downwind of the WFs and was wider in summer. The intensity and range of attenuation of WF_dw were weaker than those of WF_sz. The daytime WS in the WFs decreased by 0.2-0.3 m s^-1. The attenuation range expanded at night, and the WS dropped up to 1.0 m s^-1 in the vicinity of the WTs, with 121 grids in and around the WTs passing the significance test.

Figure 6.  Difference in mean wind speed (m s^-1) at hub height (85 m) between sensitivity and control experiments in summer. (a) Daytime and (b) nighttime.

Generally, WS decreased in the mountain WFs in both summer and winter. Because the WT thrust coefficient is set as a fixed value in the Fitch mode, WS attenuation is mainly caused by the momentum absorption and WT wake effect. The WS was reduced because the WT absorbed kinetic energy and converted it into electric energy. Therefore, the larger the background WS, the more it will decrease. Because the WS at night was greater than that during the daytime, the WS attenuation range was larger at night than that in the daytime. Because of the mountainous terrain of the study area, even though the summer background WS was slightly less than the winter background WS, there was an obvious prevailing wind direction (southwest-northeast). Wang et al. ^[38] pointed out that the seasonal variability in WS depends primarily on the prevailing wind. Therefore, the attenuation in summer was more obvious than that in winter, which was inconsistent with the results of the studies on the WFs situated in plains. The WS attenuation in summer in the plain WFs was weaker than that in winter (Li et al. ^[28]; Wang et al. ^[38]). The WS was larger and the wind direction was more stable in winter mainly due to the influence of monsoon. However, the mountainous terrain was complex and there was a SW-NE prevailing wind in summer, which made the attenuation more obvious in summer. Overall, the WS attenuation in mountain WFs was greatly affected by the terrain.

Variation in the land surface temperature (LST) in the two WFs was not consistent in winter days (Fig. 7a). The LST decreased by up to 0.06 ℃ in WF_sz, whereas it decreased by up to 0.04 ℃ in the northern WF_dw, and increased by up to 0.02 ℃ in the southern WF_dw. The LST increased in both WF_sz and WF_dw in the winter nights (Fig. 7b), and its variation was more significant than that in the daytime, increasing by up to 0.06 ℃. The LST showed no obvious variation in the daytime, and presented a slight increase in the nighttime, which did not pass the significance test.

Figure 7.  Difference in mean 2m temperature (℃) between sensitivity and control experiments in winter. (a) Daytime, and (b) nighttime.

The difference in the vertical section of temperature was obtained by making a section along the area (NE-SW, 135°) where the WFs were located. The temperature cooled from the surface to 2 km in the WFs in winter days in the vertical direction (Fig. 8a). The maximum variation is - 0.06 ℃ at approximately 600 m above the ground. The temperature mainly increased at night (Fig. 8b), where the maximum variation is more than 0.03 ℃, which was more obvious in the southern WF_dw. The temperature mainly cooled at 600 m above the ground, and the maximum decrease was above 0.03 ℃. Over the grids where the WTs were installed, the temperature mainly increased near the ground, with the most obvious variation at 50 m (Fig. 8c). The temperature increased more in the nighttime (0.038 ℃) than in the daytime (0.025 ℃). It decreased above the hub height where the coolest center was at 500 m in the daytime, and at 150 m in the nighttime; the decrease exceeded 0.03 ℃. The temperature recovered at 1200 m.

Figure 8.  Mean vertical temperature (℃) between sensitivity and control experiments in winter. (a) Daytime, (b) nighttime, and (c) over grids of wind turbines. (The wind farms' area is marked as green boxes, and the black dashed line represents the area swept by the wind turbine blades in (a) and (b); green dashed line is hub height in (c)).

The LST in the WFs varied greatly in summer than in winter, which was similar to the results of the study by Sun et al. ^[29], and it had more noisy signals, which was consistent with a previous report by Wang et al. ^[38]. In the summer daytime (Fig. 9a), the LST in the WFs showed opposite variations. In WF_sz, LST increased by 0.2 ℃, with the increase in the area surrounding the WF greater than that in the WF. The LST drop was weak in the WF_dw but became obvious around the WF, with the maximum decrease of 0.5℃. In the summer nights (Fig. 9b), LST variation in the WFs was relatively consistent, showing a weak increase of up to 0.2 ℃. These variations did not pass the significant test. There was a downdraft in summer over the WF_sz, with a vertical velocity > 0.44 m s^-1, and the WTs strengthened this downdraft. Therefore, there was an obvious warming in the WF_sz. There was an updraft over the WF_dw, which cooled the atmosphere near the ground. Therefore, the variation of LST in the mountains during summer daytime was more obviously affected by the meteorological conditions than by the WFs.

Figure 9.  Difference in mean 2m temperature (℃) between sensitivity and control experiments in summer. (a) Daytime, and (b) nighttime.

Opposite vertical temperature variations existed below 1 km in the WFs during the daytime in summer (Fig. 10a). LST mainly increased in the WF_sz, and the maximum temperature variation (0.15 ℃) was observed in the area swept by the WT blades. The cooling effect was recorded mainly above 1 km, with a maximum variation of -0.2 ℃. LST decreased in the WF_dw, with a maximum variation of - 0.2 ℃. Temperature within the area swept by the WT blades increased at night, with a maximum value of 0.10 ℃ (Fig. 10b). This increase was mainly observed over the grids where the WTs were installed (Fig. 10c), with the most obvious increase near the hub height. Temperature increase during the nighttime (0.030 ℃) was stronger than that in the daytime (0.028 ℃). Temperature decreased above the hub height (~0.04 ℃), with the coolest center at 900 m in the day and at 300 m at night. The temperature began to increase and decrease above 1300 m during the day and night, respectively.

Figure 10.  Mean vertical mean temperature (℃) between sensitivity and control experiments in summer. (a) Daytime, (b) nighttime, and (c) over grids of wind turbines (symbols are the same as those in Fig. 8).

There was an increase in the LST in the nighttime. The seasonal and diurnal variations in WS primarily determined the seasonal variations in WF-induced warming (Zhou et al. ^[39]). The WS variation in the study area was more obvious in summer, leading to more obvious LST increase in summer. The stable direction of the prevailing wind and the frequent complex synoptic-scale weather events in summer were more likely to induce strong background TKE, which masked the turbine-induced turbulence effect and consequently contributed to the noisy warming signals (Chang et al. ^[40]). The temperature firstly increased with the vertical height and then decreased. The temperature increases most obviously at hub height, which was consistent with the temperature variation in the WFs situated in plains (Xia et al. ^[25]). Unlike the WFs in the plains in North China (Li et al. ^[28]; Wang et al. ^[38]), a more obvious increase in temperature was observed during summer than in winter in the studied mountain WFs, which might be due to the effect of monsoon in the plains that had high WS in winter, and also because the direction of the prevailing wind in the mountains had a greater influence on temperature.

The surface energy field is analyzed based on the conditions at night (without net solar radiation).

The latent heat flux (LHF) in the WFs is negative in both winter and summer nights, indicating weakening of water vapor evapotranspiration. Similar results were reported by Mangara et al. ^[27] and Zhou et al. ^[39]. The LHF variation was very slight in winter (Fig. 11a) with a decrease of less than 0.8 W m^-2, whereas the variation was more obvious in summer (Fig. 11b). The WF_sz was dominated by LHF decrease, with values up to - 2.45 W m^-2. The LHF slightly increased in WF_dw, with values up to 0.6 W m^-2. The LHF variation for both summer and winter did not pass the significance test.

Figure 11.  Difference in the mean surface latent heat flux (W m^-2) between sensitivity and control experiments in nighttime. (a) Winter, and (b) summer.

Changes in surface humidity between the ground and atmosphere have a strong impact on the LHF, and are negatively correlated (Zhou et al. ^[39]). Furthermore, a decrease in WS and LST may lead to a decrease in the LHF. In winter nights, the WS in the WFs decreased (maximum ~1.1 m s^-1), LST increased slightly (maximum 0.06 ℃), and water vapor decreased (maximum 1.0 × 10^-6 kg kg^-1). Reduction in the WS and water vapor were dominant, leading to a decrease in the LHF. The range and degree of WS attenuation in the WFs in summer nights (maximum ~1.1 m s^-1) were larger than those in winter, with a more obvious warming effect in summer (maximum 0.2 ℃). Therefore, in summer nights, decrease in WS and water vapor (up to 3.0 × 10^-6 kg kg^-1) was dominant in the WF_sz, leading to a decrease in the LHF. By contrast, the warming effect and the increase in water vapor content (up to 4.0×10^-6 kg kg^-1) dominated in the WF_dw, leading to a slight increase in the LHF.

The sensible heat fluxes (SHF) are positive in both winter and summer nights in the WF, with a more obvious variation in summer. In winter (Fig. 12a), there was a slight increase in the SHF (up to 3 W m^-2), which was not very prominent in the non-WF and was consistent with the results of Xia et al. ^[42]. In summer (Fig. 12b), a total of 44 grids passed the 90% significance test, which mainly concentrated in the WF_sz (34 grids). The increase in SHF indicates that the WTs transferred more heat from the ground to the atmosphere. Furthermore, the increase in SHF was more consistent with the LST increase in spatial distribution. Previous studies suggest that SHF is the most dominant factor responsible for the simulated temperature changes at the surface (Lee et al. ^[43]). Moreover, the SHF variation decreased in the WFs of the plains (Roy et al. ^[13]; Mangara et al. ^[27]), which was slightly different from the decrease observed in our study area. According to a report by Xia et al. ^[42], this was more likely to be related to momentum sink, which significantly increased the surface heat in mountain WFs as compared with the plains.

Figure 12.  Difference in the mean surface sensible heat flux (W m^-2) between sensitivity and control experiments in the nighttime. (a) Winter, and (b) summer.

According to the parameterization principle of the Fitch model, the thrust coefficient and power curve of the the wind turbine directly affect the climate response of the wind farm (Fitch et al. ^[22]). Compared with the thrust coefficient and power curve of the wind turbine (Fig. 3), the larger the average wind speed, the smaller the thrust coefficient, the smaller the proportion of TKE absorbed by the wind turbine from the atmosphere, and the stronger the ability to produce wind power. Therefore, the near surface climate response signal generated by wind farms will be partly weakened in this case due to the simulated high wind speed.

To study the wind farms (WFs) in the mountains in northern Hubei Province (WF_sz and WF_dw), a wind farm experiment is designed with the installed capacity the same as the actual situation. Sensitivity experiments in winter (January 2014) and summer (July 2014) are conducted using the WRF-Fitch model. The impact of WFs on the local climate in mountainous areas has been studied by analyzing the differences between the sensitivity and control experiments in meteorological parameters and energy field. The following conclusions are drawn:

(1) There are different degrees of wind speed (WS) attenuation in mountainous WFs. The attenuation is greater in the nighttime than in the daytime, and greater in summer than in winter. The degree and range of attenuation are related to the background wind speed and direction of the prevailing wind.

(2) The LST in the WFs mainly increases at night, and is significantly more in summer (up to 0.2 ℃) than in winter (up to 0.06 ℃). The most obvious height of vertical temperature increase is approximately 50 m. The temperature begins to decrease as the height increases. In both winter and summer, the height of the cooling center is higher in the daytime than that in the nighttime.

(3) The LHF in the WFs decreases mainly at night, with a more obvious decrease (-2.45 W m^-2) in the WF_sz at summer nights. The SHF in the WFs increases greatly at nights in summer (up to 5 W m^-2) than in winter.

The momentum in the atmosphere decreases by the rotation of the WT blades, causing an increase in the TKE increases. This decrease is reflected in the attenuation of WS. The atmospheric stratification is unstable in the daytime and the atmosphere is uniformly mixed. Therefore, the disturbance has no obvious pattern. By contrast, the atmospheric stratification is stable at night and the surface radiation easily causes inversion. The WTs carry warm air to the lower layer, resulting in an increase in the LST. The WT operation increases TKE and changes the mode of material and energy exchange between the surface and near-surface atmosphere. The WS decreases, water vapor evaporation near the ground and water heat exchange become slow, and the LHF decreases. The water vapor content in summer is higher than that in winter, so the variation in LHF caused by changes in the water vapor is more obvious. Xia et al. ^[42] pointed out that the change in SHF is responsible for the simulated temperature change at the surface, and the momentum of the WT parameters increases the SHF signal.

The present study has analyzed the effect of wind farms on the local climate in the mountainous areas and compared it with those in the plains. However, the results obtained are based only on small-scale WFs in southern China. The plain WFs used for comparison are larger and mainly located in northern China. In addition, the climate varies between different regions. These factors may also be responsible for the difference observed between the climate in the plains and the mountains, which is a problem that can be studied in future work. Furthermore, different underlying surface conditions, the layout of wind turbines, and hub height all have an impact on the influence of wind farms on local climate. More cases are needed to study the impact of wind farms on weather processes, atmospheric boundary layer characteristics, and climate effects, which all play important roles in the comprehensive understanding the effect of wind farms on climate.