Ground-level climate at a peatland wind farm in Scotland is affected by wind turbine operation

This research was undertaken at Black Law Wind Farm, Scotland (55°46'01''N 03°44'20''W, elevation 250–320 m), which comprises 54 turbines within 18.6 km². The turbine blade hub heights are approximately 70 m, the rotor diameter 82 m and the total capacity is 124 MW. The wind farm was operational (hereafter referred to as ON) from 24th May 2012 to 7th June 2012, idle (hereafter OFF) from 12th June 2012 to 25th July 2012 and operational (ON) from 28th July to 15th November 2012. The switching on and off of the turbines was a phased operation lasting several days; data from this period were excluded from analysis. Given the commercial sensitivity of the data and the de-powering of the entire site, no wind data at hub height was available from which to calculate the capacity factor. However, the average capacity factor (i.e. the ratio of actual output compared to the maximum potential output) in the UK in 2012 was 26.2% (DECC 2014).

The vegetation across the wind farm comprised coniferous plantation (tree height up to approximately 20 m) and blanket bog, with a small amount of acid and improved grassland. However, most monitoring was undertaken in the blanket bog, with a limited number of sensors in grassland areas. The topography of the site was relatively flat, with most of the site lying between 260 and 320 m and with the topography varying predominantly in the north–south orientation (figure 1).

Zoom In Zoom Out Reset image size

Our field data collection framework was designed to capture the impacts on ecosystem processes within a wind farm, in terms of the scale, density and resolution, and did not undertake an energy balance approach. Air temperature (T_A) and relative humidity (RH) were measured every second and five-minute averages recorded at 2 m above the land surface (HOBO U23-002, Onset, USA; see SI). Sensors were deployed in a grid at 101 locations across a 2.6 by 1.4 km area of Black Law Wind Farm (figure 1). The elevation of the sampling locations varied from 277 to 305 m (no elevation corrections were required; see SI for details). Sensor resolution was 0.02 °C at 25 °C for temperature and 0.03% for RH. Surface and soil (−5 cm) temperatures (T_SU and T_SO respectively) were recorded every 30 min (HOBO Pendants, Onset, USA) at 36 locations, clustered at four sites (figure 1). Sensor resolution was 0.14 °C at 25 °C. At the same sites soil moisture (−10 cm) was measured every minute and averaged every 30 min (using a site-specific calibration) (CS625 water content reflectometers and CR200 loggers, Campbell Scientific Limited, UK). Sensor resolution was 0.1% volumetric water content. Wind direction was measured with a 2D sonic anemometer (Gill Instruments, UK) at site B, 2 m above the land surface every 10 s and averaged over 10 min intervals (figure 1). Wind speed resolution was 0.01 m s⁻¹ and direction was 1°. All sensors were checked and calibrated prior to deployment. The sampling strategy was chosen to allow an adequate representation of the response of microclimate to wind turbine operation at a resolution appropriate for ecosystem processes (see SI). To ensure the period of field measurement was 'normal' we examined wind speed and stability and mixing ratios using data from nearby Met Office stations (see SI for full details).

Sunrise and sunset are the meteorologically relevant temporal controls on boundary layer development (Stull 1998). Consequently, 'day' was classified as one hour after sunrise to one hour before sunset, as defined by the National Oceanic and Atmospheric Administration algorithm (ESRL 2014), and 'night' as one hour after sunset to one hour before sunrise (transition periods were excluded). As the Sun rose and set at significantly different times during the period of measurement, fraction of day was calculated, with sunrise as 0.0 (or 1.0), sunset as 0.5, and time linearly scaled between. The fraction of the day data were then categorised into 24 pseudo-hourly bins.

AH was derived from RH by first calculating the saturated water vapour pressure, then calculating the vapour pressure, and then finally the AH based upon temperature and vapour pressure (see SI for more information).

Departures were used in the analysis of the T_A, AH and RH data to remove diurnal and seasonal signals (the greatest controls) from the data. Essentially, the site-wide instantaneous mean was calculated for all measurement locations, and subtracted from each individual measurement, and then averaged over the ON and OFF periods and for downwind and not downwind groups where appropriate. In numerical form, the departures were derived by considering the vector of N temperature measurement locations at time t:

$$\begin{eqnarray}\underline{T}(t) & = & [{T}_{1}(t),{T}_{2}(t),\ldots .,{T}_{N}(t)]\;{\rm{for}}\;t={t}_{1},\\ t & = & {t}_{2},\ldots ,t={t}_{M}\\ & & ({\rm{i}}.{\rm{e}}.\;M\;{\rm{times}}\;{\rm{in}}\;{\rm{the}}\;{\rm{sample}}\;{\rm{set}})\end{eqnarray} \tag{ 1 }$$

then, calculating the site-wide scalar mean temperature at time t = t_j

$$\begin{eqnarray}&&{T}_{{\rm{AV}}}({t}_{j})=\tfrac{1}{N}\displaystyle \sum _{i=1}^{N}{T}_{i}({t}_{j}).\end{eqnarray} \tag{ 2 }$$

The vector of departures from the mean at time ${t}_{j}\;$

$$\begin{eqnarray}&&\underline{T}\text{'}({t}_{j})=[{T}_{1}({t}_{j})-{T}_{{\rm{AV}}}({t}_{j}),{T}_{2}({t}_{j})-{T}_{{\rm{AV}}}({t}_{j}),\ldots ,\\ &&\quad \quad \quad {T}_{N}({t}_{j})-{T}_{{\rm{AV}}}({t}_{j})]\end{eqnarray} \tag{ 3 }$$

allowing the time-averaged mean departure vector to be constructed:

$$\begin{eqnarray}&&\bar{\underline{T}}^{\prime} =\left[\tfrac{1}{M}\displaystyle \sum _{j=1}^{M}{T}_{1}^{\prime }({t}_{j}),\tfrac{1}{M}\displaystyle \sum _{j=1}^{M}{T}_{2}^{\prime }({t}_{j}),\;\ldots ,\right.\\ &&\quad \quad \left.\tfrac{1}{M}\displaystyle \sum _{j=1}^{M}{T}_{N}^{\prime }({t}_{j})\right].\end{eqnarray} \tag{ 4 }$$

This vector is the one used in the paper, and when departures are mentioned they refer to this vector, or elements of it.

Variation in T_A, T_SU, T_SO, and AH was assessed by calculating the coefficient of variation (c_v) across the site for each measurement interval, and averaging in pseudo-hourly bins (based on fraction of the day) for ON and OFF periods (figure 3). Given the c_v of a population is defined as the standard deviation divided by the mean of the population (c_v = σ/μ), the temperatures were first converted to degrees Kelvin to ensure meaningless values of c_v were not generated from negative or zero values of μ. The c_v was calculated using the same steps as above for T_A, AH and RH departures: the site-wide c_v was calculated and the time-averaged c_v vector constructed. In numerical form, at time t = t_j

$$\begin{eqnarray}&&{c}_{{\rm{v}}}({t}_{j})=\;\displaystyle \frac{\tfrac{1}{N}\displaystyle {\sum }_{1}^{N}{({T}_{i}({t}_{j})-{T}_{{\rm{AV}}}({t}_{j}))}^{2}}{{T}_{{\rm{AV}}}({t}_{j})},\end{eqnarray} \tag{ 5 }$$

where the temperature vector T and site-wide average T_AV are defined above (equations (1) and (2)). Analogous formulae apply to the other fields considered (AH and surface and soil temperatures). Note that the c_v was subsequently binned, according to time of day, giving a measure of spatial and temporal variability.

To examine the magnitude of effect of the turbines on T_A, RH and AH, only data when the wind direction at 2 m was from between 220° and 240° were considered. This directional sector was chosen as it was aligned with the main axis of the wind farm (figure 1), thus any cumulative effect of the turning turbines was likely to be greatest, and it was approximately the dominant wind direction. For the data to be included the wind had to be originating from between 220° and 240° at the measurement time and the previous 30 min, and also satisfy the day–night criteria for the entire 30 min period. This allowed the development of any underlying signal to be captured, while still providing a significant amount of data. Departures for all the data that fit these criteria were calculated for each sampling location, (see above), thus allowing the relative T_A, RH and AH of sites classified as downwind and not downwind of turbines to be compared. Further, as no hub height wind data was available, to take into account the potential effects of directional shear (between surface and hub height), we repeated the analysis for 200°–220°, 220°–240° and 280°–300° wind direction sectors. Unfortunately, given the layout of the wind farm, all measurement sites were classified as downwind in the 240°–260° and 260°–280° direction sectors. Given the complexity of wake dynamics, especially the interaction of multiple wakes, it was not appropriate to make assumptions regarding wake expansion and movement. Thus our analysis is prudent: measurement locations categorised as downwind (not downwind) may have been not downwind (downwind) of turbines and consequently the differences in T_A, RH and AH we quantified may be smaller than actual.

Logarithmic best-fit lines were used to characterise the relationship between temperature and AH departure and distance to the nearest turbine (figure 2). Differences in c_v between ON and OFF periods for each pseudo-hour (figure 3) and temperature departures between downwind and not downwind groups were tested (table 1) using a paired t-test with unequal variances using Stata13 (StataCorp, Texas).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To determine the integrated effect (i.e. irrespective of wind direction and speed) of the whole wind farm across the measurement period, we analysed the mean day and night-time differences in T_A and AH departure for each site from the site-wide mean between ON and OFF periods, using all data (i.e. data were not categorised as downwind or not downwind and therefore wake dynamics do not need to be considered). To assess the spatial extent of effects we related the departures to distance from the nearest turbine.

During the night, air closer to a wind turbine was warmer and more moist, with T_A departures reaching 0.25 °C and AH departures 0.1 g m⁻³ (figure 2(a)). Night-time warming and moistening has been attributed to downward mixing of warmer moister air by turbines during stable conditions (Baidya Roy 2011). The potential occurrence of this at Black Law is supported by analysis of midnight soundings from the two nearest upper-air stations: this reveals that during the most stable conditions the lapse rate is positive for both temperature and moisture (see SI) (Baidya Roy 2011). The larger effects closer to wind turbines could be the composite effect of that wind turbine and others upwind. This suggests that the interactions between multiple wakes may enhance effects found in single row turbine studies (Rajewski et al 2014). While positive departures are evident up to 200 m (2.4 rotor diameters) away (figure 2(a)), this does not suggest that effects of downward mixing were only evident within 200 m (2.4 rotor diameters) of a turbine. Compared to the site-wide mean, it was generally warmer and more moist within 200 m (2.4 rotor diameters) of a turbine and cooler and drier beyond 200 m (2.4 rotor diameters). During the day, air closer to a wind turbine was cooler, with departures up to 0.05 °C, but AH was not influenced (figure 2(b)). This weaker day-time effect could be attributable to a convectively driven, well mixed boundary layer (Zhou et al 2013).

The trends between T_A departure (ΔT_A = T_A[ON] − T_A[OFF]) during both day and night, and AH departure (ΔAH = AH[ON] − AH [OFF]) at night, relative to distance from the nearest turbine (x), can be approximated by logarithmic functions: ΔT_A = 0.62 − 0.12 ln (x), r = 0.75 and ΔAH = 0.15 − 0.03 ln (x), r = 0.61 during the night and ΔT_A = −0.11 + 0.02 ln (x), r = 0.53 during the day (figure 2). The daytime ΔAH cannot be approximated by a logarithmic function (r < 0.2). These logarithmic trends demonstrate that the effect of wind turbines can be quantified and thus potentially represented in models of Earth surface energy balance. This would be valuable given the complexity of extrapolating impacts from studies examining single or a single row of turbines (Rajewski et al 2014). Currently, model parametrisations (for example, Fitch et al (2012)) typically represent the effect of turbines on the atmosphere by imposing a sink of momentum and a source of turbulence kinetic energy. Multiple turbines are represented by integrating over a typical model grid cell (which may contain more than one turbine). With all such parametrisations a number of coefficients are required (such as the fraction of energy converted into turbulent kinetic energy—unknown a priori—which controls the subsequent mixing). Thus, the integrated effect we identified could be used to test such parameterisations of wind farms in numerical weather prediction models (Cervarich et al 2013), where the spatial resolution is greater than the distance between individual turbines, and the integrated effect of a whole wind farm needs to be established.

To quantify the spatio-temporal effects of wind turbine operation we examined the variation in T_A, T_SU, T_SO, and AH throughout the diurnal cycle (irrespective of wind direction and thus wake dynamics do not need to be considered). We used the coefficient of variation c_v of each of the measurements across the whole site for each measurement interval averaged for both the ON and OFF periods as a function of time of day. Using the c_v allows comparison of the ON and OFF periods: reporting this in degrees (such as would be obtained by the standard deviation) would be misleading as the temperature is more variable in summer months compared with autumn months.

The spatial variations in the T_A, T_SU, T_SO and AH data were significantly greater during the ON period compared with the OFF period (p < 0.05 for pseudo 10:00 and 11:00 for the T_SO data and pseudo 07:00 for the AH data, p < 0.01 for all other data and hour intervals) (figure 3), suggesting turbine operation increased vertical mixing and turbulence. This effect, at several levels (i.e. soil, surface, and air at 2 m), has not been identified in other field studies but has been reported in modelling studies (Lu and Porté-Agel 2015). The spatial variance in microclimate is of crucial importance as ecosystem processes respond to small-scale variation in climate (Baidya Roy et al 2004, Baidya Roy and Traiteur 2010, Baidya Roy 2011, Zhou et al 2012, De Frenne et al 2013, Rajewski et al 2013, Smith et al 2013, Zhou et al 2013). The differences in temperature variation between ON and OFF periods were greatest for T_SU data (figure 3(b)) and smallest for T_A data (figure 3(a)), reflecting relatively well-mixed air at 2 m and increased variability at the surface arising from peatland micro-topography and vegetation shading. AH aside, the spatial variance was greatest for T_SU and least for T_A, during ON and OFF periods (figure 3). This is attributable to the patchy influence of vegetation shading and surface water content on the temperatures of the surface sensors. Further, peatland temperatures are highly spatially variable, in response to the peat thermal properties (Kettridge and Baird 2010), thus explaining the T_SO spatial variability. In contrast, although there are differences across the wind farm, the air max is relatively well mixed.

The difference in T_A and AH variability between the ON and OFF periods was greater during night than day (figures 3(a) and (d)), whereas it was approximately equal for T_SU and T_SO, with smaller differences during transition periods around sunrise and sunset (figures 3(b) and (c)). This suggests that night-time T_A and AH were most sensitive to turbines, due to downward mixing of warm air. Further, the mixing down of warm air during the ON period appears to have affected the diurnal trend in T_SO and AH: the maximum variability in T_SO and AH were later in the day during the OFF period compared with the ON period (figures 3(c) and (d)). This suggests that the night-time background gradient of T_SO and AH are eroded by turbulence earlier in the day during the ON period. No trends in soil moisture were found (see SI) probably as a result of the high water table across the wind farm (Armstrong et al 2015).

Our final analysis examined the magnitude of effect of wind turbines on T_A and AH during the night (see SI for day-time and RH results) when we observed a strong relationship between T_A and AH departures and distance from the turbine (figure 2). Consequently, we excluded all data from one hour before sunrise to one hour after sunset and filtered for wind directions between 220° and 240° (aligned along the main axis of the wind farm) (figure 1). Temperature departures for all the data that fit these criteria were calculated for each sampling location, thus allowing the relative T_A and AH of sites downwind and not downwind of turbines to be compared. Directional shear between hub height and the surface, both clockwise and anticlockwise, can occur (Walter et al 2009, Cariou et al 2010, Rhodes and Lundquist 2013), and consequently we also analysed the data filtered for wind directions between 200°–220°, 220°–240° and 280°–300°.

We found that temperatures were significantly warmer in areas downwind of turbines during the ON period (p < 0.01), with an average relative warming of 0.18 °C for the 200°–220° direction sector and 0.16 °C for the 220°–240° sector (table 1). The AH of air downwind of the turbines was, on average, 0.03 g m⁻³ greater during the ON period (p < 0.01) for the 200°–220° direction sector but there was no significant difference for the 220°–240° sector (table 1). This suggests that there may have been slight anti-clockwise directional shear in the boundary layer, although we cannot rule out stronger slightly clockwise shear as all sites were downwind of turbines in the 240°–260° and 260°–280° sectors. Although clockwise shear is expected given the Ekman spiral, anti-clockwise shear has been observed previously (Walter et al 2009) and could have been caused by the wind turbines disturbing the atmospheric boundary layer.

Analysis of the data shows that the relative increase in saturated water vapour pressure had more effect on AH than the combined increase in temperature, and lowering of RH (see SI), consistent with turbine-induced mixing of warmer and more moist air downwards (Baidya Roy et al 2004). During the OFF period, temperature and AH departures were variable with no statistically significant difference between downwind and not downwind sites (p > 0. 05), further demonstrating the variation during the ON period was due to wind turbine operation (table 1).

Terrestrial carbon cycling is highly variable, both spatially and temporally, and influenced by many biotic and abiotic conditions and their interactions, as demonstrated at this site (Armstrong et al 2015). Given this, and the relatively small variation in microclimate between ON and OFF periods compared to spatial and temporal variability, it was not possible to measure the impact of wind farm operation on carbon cycling. However, peatlands are highly temperature sensitive environments, and consequently these observed small-scale changes in temperature could have significant implications for peatland carbon stocks. For example, during the growing season, a 0.88 °C warming was found to increase rates of ecosystem respiration in a peatland in northern England by 20% and decrease net CO₂ uptake by 11% (Ward et al 2013). In contrast, during the non-growing season when CO₂ fluxes were much lower, a 0.72 °C warming increased ecosystem respiration rates by 44% and CO₂ uptake by 7% (Ward et al 2013). Therefore, as we found night time warming was most influenced, decomposition processes may be accelerated and thus soil carbon losses observed. However, effects on respiration could also be offset by plant physiological responses to warming (Peng et al 2013). Our results also indicated that wind farm operation had greater effects on spatial and diurnal variability in T_SU and T_SO than T_A. T_SU and T_SO are recognised as stronger regulators of plant-soil carbon dynamics (Graae et al 2012, De Frenne et al 2013); consequently, the effects on the net carbon balance of the hosting ecosystem may be stronger than inferred from T_A (Baidya Roy and Traiteur 2010, Rajewski et al 2013, Smith et al 2013).