Do large-scale wind farms affect air quality forecast? Modeling evidence in Northern China

Wind farms have been found to alter local and regional meteorology and climate. Here, we show that multiple large-scale wind farms might disturb air quality forecasts and affect PM2.5 air pollution. We explore the impact of large-scale wind farms on PM2.5 concentrations and forecasts in the Northern 15 China Plain in winter and summer using a coupled weather forecast atmospheric chemistry model (WRF-Chem). Modelling results reveal that the large-scale wind farms decrease PM2.5 levels within the wind farms and increase PM2.5 concentrations by 49% and 16% of the modelled monthly mean PM2.5 concentrations in proximate areas and regions hundreds of kilometres downstream. The wind farm-forced changes in PM2.5 are more evident in the simulated hourly PM2.5 concentrations. The model sensitivity 20 studies reveal that hourly concentration fractions in winter induced by wind farms vary from -40% to 250% in nearby and distant downstream regions and metropolises, comparing with the cases without the wind farms. The impact of wind farms on modeled PM2.5 during the nighttime is stronger than that in the daytime. Our results suggested that the wind farm perturbed changes in PM2.5 should not be overlooked because such changes might affect air quality forecast on an hourly basis, particularly in heavily 25 contaminated Beijing-Tianjin-Hebei region by PM2.5.

roughness (Mo et al., 2017;Keith et al., 2004;Ivanova and Nadyozhina, 2000). This parameterization scheme calculates the aerodynamic roughness length Z0 (m) of a wind farm, defined by as (Lettau, 1969): where h * is the average vertical extent (m) or hub height for wind turbines and s is the silhouette area (km 2 ) of the average obstacle or the rotating area for wind turbine blades. S is the density of roughness 130 elements, calculated by S=A/n, where A is the total area occupied by obstacles (km 2 ) and n the total number of obstacles. As aforementioned, the wind turbine spacing is set as six times the diameter of the wind turbine rotor. By setting s and S as 0.01 km 2 and 0.46 km 2 , we obtained the roughness length Z0 of 1.04 m. This method has the advantages of simplicity and accuracy in representing the wind turbines (Wieringa, 1993;Petersen, 1997).

135
The second scheme, developed by Fitch et al. (2012), considers the wind turbines as a momentum sink within the turbine rotor area, which transforms kinetic energy (KE) into electricity and turbulent kinetic energy (TKE, m 2 s -2 ). This scheme extracts the total fraction of KE from the air utilizing a thrust coefficient CT, which is turbine type dependent and the function of wind speed. A proportion KE is converted to electrical energy by a power coefficient CP. The rest of KE is converted to TKE, defined by 140 the TKE coefficient CTKE=CT-CP. The CT and CP can be taken from the turbine manufacturers. The WRF model implements the Fitch scheme (Fich et al., 2012(Fich et al., , 2015Yuan et al., 2017) which accounts for the effects of local wind drag on wind-energy extraction and power estimation (Lee and Lundquist, 2017).
This scheme has been extensively applied in windfarm-meteorology interactive modelling using WRF model (Cervarich et al., 2013;Xia et al., 2016Xia et al., , 2017Fiedler and Bukovsky, 2011;Vautard et al., 2013; https://doi.org/10.5194/acp-2019-991 Preprint. Discussion started: 25 March 2020 c Author(s) 2020. CC BY 4.0 License. Mo et al., 2017;Yuan et al., 2017;and Sun et al., 2018), and was also adopted here with a standing minimal thrust coefficient of 0.16. However, Volker et al. (2012) have reported that Fitch-scheme estimated thrust applied to the flow was overestimated by almost one order of magnitude and that the modeled TKE in the Fitch-scheme diffused the velocity deficit deep into the boundary layer, and caused unnaturally high positive velocity deficits at the lower boundary.

150
If the turbines are assumed to be oriented perpendicular to the wind flow, the drag force created by wind turbines is defined by: where  is the air density, A=(/4)D 2 is the cross sectional rotor area (D is the diameter of the turbine blades), and V(u, v) is the horizontal velocity.

155
The rate of loss for KE from the air subject to a wind turbine reads: The electric energy extracted from KE is defined by: and the TKE extracted from the rest of KE is given by: where the nominal power of a turbine is taken as 1.5 MW.
To examine the sensitivities of air quality in the NCP to WFC, we conducted extensive model simulations of PM2.5 in the NCP subject to four model scenarios ( Table S2). The first model scenario is the control run (scenario 1, S1), in which the WFC was not taken into consideration, hereafter referred 165 to as the BASE or no-WFC simulations. The second and third model scenarios used the surface roughness length parameterization scheme (referred to as the SRL run, S2) and the drag force parameterization (referred to as the DFP run, S3). Considering the future wind farm development in northern China, in the fourth model scenario we projected a double area of WFC with the drag force parameterization from the current wind farm area as shown in Figure 1 (referred to as the DOU run, S4). WRF-Chem was run for 170 January and July 2016 to examine the responses of PM2.5 to different setups of the model scenarios under typical winter and summer atmospheric circulations.

WFC disturbed hourly and daily PM2.5 in winter
Figure S10a shows WRF-Chem-simulated monthly averaged daily air concentrations of PM2.5 from the BASE run (S1, Methods) at the lowest model level (~5 m above the ground surface) across the model 185 domain in January 2016. High concentrations can be observed in the NCP due to strong emissions of PM2.5 precursors (e.g., sulphur dioxide, SO2, and nitrogen oxide, NOx) in these heavy industrial regions featured by steel, energy, and cement industries (Zhao et al., 2013;Xu et al., 2016;Wang et al., 2018). Province from all three model scenario runs. The maximum PM2.5 difference was as high as 14 µg m -3 in the highly polluted Beijing-Tianjin-Hebei region, which was a maximum 10% increase in monthly mean PM2.5 concentration. This is also illustrated by the fraction of mean PM2.5 concentrations from the SRL (S2) and DFP (S3) runs to that from the BASE run (S1) in Figure 2a   precursors. In this sense, the regional wind field plays a more important role than the WFC-perturbed wind field.  (Garratt, 1994). Comparisons of the modelled monthly mean wind speed differences between three WFC runs and the BASE run in Figure S13 show smaller areas of negative mean wind differences in July than January, indicating that the wind deficits induced by the WFC in July is less significant than January. Differing from TKE as shown in Figure 4e  Hourly and monthly concentration fractions of the SRL run to that of the control (BASE) run in the selected five regions and cities are illustrated in Figure 6. Compared with the winter case (Figure 3a), the hourly concentration fractions do not significantly fluctuate (Figure 6a), which suggests that during 330 most of July, the WFC did not frequently disturb hourly changes in PM2.5 concentrations. However, we can still identify strong oscillations, with the maximum amplitude as high as 400% in the hourly PM2.5 concentrations fractions at 2100 LST July 15, 2016 in NHB and 1400 LST July 24, 2016 in ZJK, which is stronger than that in the wintertime. The monthly averaged concentration fractions are all positive in the five selected regions and cities (Figure 6b), with the highest fraction of 11.9% in ZJK, followed by 335 10.5% in NHB, 5.1% in Tianjin, 5% in Beijing, and 4.3% in CHB, which again indicates that the WFC increases PM2.5 concentrations in these places. As shown in Figure 6b, the prevailing wind in northern China in summer consists of mainly southerly and southeasterly winds under the East Asian summer monsoon regime. The wind direction alters more frequently in summer than in winter. The average wind deficit within the WFC is approximately 2 m s -1 and 1 m s -1 around the WFC. These wind speed deficits 340 are much smaller than that in winter but are still visible, which likely results in weak fluctuations of the modelled PM2.5 concentrations.

Daytime and nighttime PM2.5 perturbed by WFC
Baidya Roy and Traiteur found that a wind farm exerted a significant impact on atmospheric stability and the development of the boundary layer over the wind farm, with obvious diurnal variations (Baidya      simulations; (c) same as Figure 5b but for the differences between DFP (S3) and BASE (S1) simulations; (d) same as Figure 5b but for DOU (S4) and BASE (S1) simulations. PM2.5 differences are calculated by (CSi -CBASE), where CSi denotes modelled concentrations from different model scenarios (i=2, 3, 4).
The areas where the monthly PM2.5 fractions are significant at the 95% confidence level (t-test) are highlighted by the black dots.