Micrometeorological impacts of offshore wind farms as seen in observations and simulations

In Europe, offshore wind farms have a capacity of 16 GW, with 71% installed at the North Sea. These wind farms represent an additional source of turbulence and may influence the stratification of the marine boundary layer. We present aircraft measurements and simulations showing an impact on temperature and humidity at hub height in the order of 0.5 K and 0.5 g kg−1 even 60 km downwind of a wind farm cluster. We extend these simulations to explore a realistic future scenario, suggesting wakes in potential temperature and water vapor propagating more than 100 km downwind. Such impacts of wind farms are only observed in case of a strong stable stratification at rotor height, allowing wind farms to mix warmer air downward.


Introduction
The offshore wind energy market grew rapidly in the year 2017-compared to the year 2016, 2.6 times more wind farms were installed offshore in Europe. Almost 85% of these wind farms were installed in the North Sea (WindEurope 2017). These investments are motivated by the stronger and steadier winds in the North Sea compared to onshore sites (Bilgili et al 2011) as well as relatively shallow water depth in the North Sea (WindEurope 2017).
On-and offshore wind farms can affect the micrometeorology of the boundary layer. Wakes generated by single wind turbines reduce momentum downwind, resulting in a wind speed deficit (e.g. Lissaman 1979, Barthelmie et al 2010, Hirth and Schroeder 2013, Rhodes and Lundquist 2013, Djath et al 2018. Christiansen and Hasager (2005) observed offshore wakes via sythetic aperture radar satellite images and showed that these wakes can propagate 20 km downwind. These results were confirmed, recently, by airborne measurements taken downwind of large offshore wind farms at the North Sea (Platis et al 2018). During stable atmospheric conditions, these offshore wakes can be longer than 70 km at offshore sites (Platis et al 2018).
Onshore wind farms can impact the boundary layer (e.g. Baidya Roy and Traiteur 2010, Zhou et al 2012, Rajewski et al 2013, Smith et al 2013, Rajewski et al 2014, Armstrong et al 2016. For example, Zhou et al (2012) observed a warming of 0.5 K in the vicinity of onshore wind farms, especially during nocturnal stable conditions. Therefore, the implications of onshore wind farms on agriculture are discussed (e.g. Baidya Roy and Traiteur 2010, Rajewski et al 2013, Smith et al 2013, Zhang et al 2013. Most offshore observational studies have so far generally focused on the wind and power deficit observed in and downwind of large wind farms (e.g. Barthelmie et al 2010, Nygaard 2014, Nygaard and Hansen 2016 except Foreman et al (2017). Only few studies have investigated the potential effect of wind Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence.
Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. farms on the marine boundary layer (MBL). These studies were motivated by visible cloud effects as they were seen in photos taken at a wind farm at the coast of Denmark (Emeis 2010, Hasager et al 2013, indicating fog formation and dispersion due to enhanced mixing downwind of wind farms. Associated with enhanced mixing, Foreman et al (2017) reported a decreased sensible heat flux downwind of a small offshore wind farm during stable conditions in the North Sea.
Modeling studies suggest a change in temperature and moisture downwind of offshore wind farms. Vautard et al (2014) obtained increased temperatures at the North Sea in the area of offshore wind farms in their simulations. In contrast, Wang and Prinn (2011) and Huang and Hall (2015) reported a potential cooling effect in the vicinity of offshore wind farms due to an increased latent heat flux, although Wang and Prinn (2011) represented the offshore wind farms as areas of increased surface roughness. Fitch et al (2013) showed that this roughness approach is not suitable to investigate climate impacts. However, no field measurements have so far investigated this potential cooling or warming effect.
Herein, we present aircraft observations accompanied with mesoscale simulations to provide a first look into spatial dimensions of these important impacts. Within the research project WIPAFF (Emeis et al 2016), 26flights were conducted in the far field of wind farm clusters. During nine flights we observed differences between potential temperature and/or water vapor concentration within and outside of the wake. Herein, we present one case study in detail and compare this case to the remaining 25 aircraft measurements to identify atmospheric conditions that favor thermodynamic impacts of wind farms on the MBL. We present these impacts to constrain maximal possible impacts of offshore wind farms on the MBL. We want to answer the following questions: • How large are the micrometeorological impacts of offshore wind farms on the MBL?
• How is the micrometeorological impact caused?
• What is the influence of existing and planned offshore wind farms on the MBL at the North Sea?
In section 2, we present the model configuration and the aircraft observations used to determine the impacts of wind farms on the MBL. In section 3 we compare the observation to our simulations up-and downwind of the wind farms. A discussion in section 4.1 compares cases with impacts on temperature with cases having no impact. Further, we suggest vertical profiles under which an impact on the MBL can be expected. Section 4.2 addresses the implications of our results with respect to existing and planned offshore wind farms. This study ends with a conclusion (section 5).

Numerical simulations
The numerical setup and the parameterizations in this study are the same as in Siedersleben et al (2018); the reader is referred to this paper for full details. Numerical simulations are performed with Weather Research and Forecasting model WRF3.8.1 (Skamarock et al 2008) with three domains having a resolution of 15, 5 and 1.67 km (figure 1), resulting in up to six wind turbines per grid cell. ECMWF analysis data provide the initial and the boundary condition for the simulation. The model uses 50vertical levels, with a spacing of ≈35 m at the bottom with the lowest level at 17 m above mean sea level, resulting in one level below rotor height and three levels intersecting with the rotor area (figure 2).
We use the wind farm parameterization (WFP) of Fitch et al (2012) to simulate the interaction between atmosphere and wind farms. The WFP extracts kinetic energy from the mean flow and acts as a source of turbulence, depending on the thrust and power coefficients of the installed turbines in the model (Fitch et al 2012, Jiménez et al 2015, Lee and Lundquist 2017. Therefore, the WFP interacts with the planetary boundary layer scheme of Nakanishi and Niino (2004) that we use in all three domains.
We focus mainly on wakes generated by a wind farm cluster consisting of the wind farms Meerwind-Sued|Ost, Nordsee Ost and AmrumbankWest (see close-up, figure 1). The observations conducted on 10 September 2016 were carried out at this wind farm cluster. These three wind farms have two turbine types: SIEMENS SWT 3.6-120 and SENVION6.2, with 90 and 95 m hub heights and rotor diameters of 120 and 126 m. The thrust and power coefficients of these wind farms are not available to the public. Therefore, we use the coefficients of the wind turbine Siemens SWT3.6-120-onshore as these are available (http://wind-turbine-models.com/ turbines/646-siemens-swt-3.6-120-shore (30 October 2018)). We have shown in Siedersleben et al (2018) that the errors introduced by these uncertainties have only a marginal effect on the wake effect.
Additional simulations include all approved offshore wind farms under construction at the NorthSea (i.e. all blue and orange wind farms shown in figure 1). For these simulations we assume the same wind turbine type for simplicity, the SIEMENS SWT 3.6-120. To assess the overall impact of these wind farms at the North Sea, we conducted a second simulation with the WFP switched off. We refer to the simulations without wind farms as no wind farm simulation (NWF) and to the ones using the WFP as wind farm simulation (WF).

Aircraft measurements
As part of the research project Wind Park Far Field (WIPAFF) (Emeis et al 2016), 26flights were conducted in the far field of large offshore wind farm clusters at the North Sea from September 2016 to October2017 (Platis et al 2018) (see table 1 for a overview). During eight flights in this project we observed a change in temperature and/or humidity within the wake of a wind farm cluster; one case is difficult to interpret and will be discussed in section 4. We will focus on aircraft measurements conducted on 10September 2016 from 08:00 to 11:00UTC and then compare the results against 25 other flights.
On 10 September 2016, the aircraft (Dornier DO-128 of TU Braunschweig) sampled the wind vector, humidity, pressure and temperature at 100Hz Besides the measurements at hub height downwind, the aircraft conducted measurements along the vertical cross-section indicated with A, B in figure 1. The aircraft flew at five different heights, starting at 60 m AMSL followed by measurements at 90, 120, 150, and 220 m AMSL along the vertical A-B. With these measurements, we investigate the effect of large offshore wind farms on the stratification of the MBL. The measurements along vertical A-B started at 10:00UTC and ended at 11:00UTC.
The aircraft probed the atmosphere 5 km upwind of the wind farm cluster to quantify the upwind conditions at 08:00UTC (see location in figure 1).

Impacts on temperature and water vapor on 10 September 2016
First the model is compared upwind of the wind farms to the aircraft measurements (section 3.1) before the impact of wind farms on temperature and water vapor is evaluated in sections 3.2 and 3.3.

Stratification of the atmosphere upwind of the wind farm cluster
The atmosphere upwind of the wind farms was stably stratified at 08:00UTC 10September 2016 (figures 2(a)-(c)). From hub-height at 90 m to the top of the rotor Table 1. Overview of flights conducted within the WIPAFF project downwind of large offshore wind farms. The numbering of the aircraft measurements corresponds to the numbering in figures 5 and 6. The letters A and G indicate the measurement location; A refers the to wind farm cluster consisting of Amrumbank West, Meerwind Sued|Ost and NordseeOst and G for the wind farm Godewind. The locations of these wind farms are indicated in figure 1. The column indicated with wsp, shows the measured wind speed at hub height according to figure 5. The sixth and seventh column indicate whether the wind farms had an impact on temperature or humidity at hub height downwind. The atmospheric stability during each measurement is shown in the last column according to the potential temperature gradient within the rotor area shown in figure 5. Observations where wind farms had an impact on the atmosphere are listed at the beginning of the

Temperature
Behind the wind farms, warmer air was observed and simulated at hub height within the wake even 45 km downwind of the last turbine (figures 3(a) and (c)).
According to the simulations, potential temperature in the wake is 0.4 K warmer than the air upwind. This effect is more pronounced in the observations. At hub height in the upwind climb flight, a potential temperature of 291.2 K was measured (figure 2) compared to maximal 291.8 K within the wake, indicating a warming of up to 0.6 K at hub height. Additionally, the observations show a stronger horizontal difference between wake and no-wake region downwind of the wind farm cluster. At the eastern flank a potential temperature gradient of 0.8 K was observed, in contrast to a difference of 0.4 K in the simulations. The model has a cold bias of ≈0.6 K compared to the observations. However, the model is stably-stratified above hub height, corresponding to the observations (more details in Siedersleben et al (2018)). As we want to investigate the impact of wind farms on the MBL and not the bias of the simulations, we add 0.6 K to all shown potential temperature figures in this study.
The potential temperature wake was associated with a mixed layer, as seen in the cross-section 5 km downwind of the wind farm cluster (figures 4(a) and (c)). However, both cross-sections from observation and simulation indicate that warmer air was mixed downward. The mixed layer extends up to 120 m AMSL in the observations ( figure 4(a)); in the simulation this neutral layer is only 100 m thick.
Mixing warmer air downward corresponds to an enhanced sensible heat flux downward ( figure 4(e)). In the observations, the atmosphere was stably stratified, consequently, we expect a sensible heat flux towards the surface (i.e. a negative sensible heat flux). Figure 4(e) shows the difference in sensible heat flux between a WF and a NWF simulation along the crosssection C-D. The blue contours indicate that the wind farms caused a greater downward heat flux above and within the farm, hence, explaining the warming below ≈180 mAMSL (figure 4(e)). The simulations show cooling aloft right above the farm area and warming within the farms but starting half-way through the farm area and extending much farther downwind than the cooler area figure (4(f)).

Water vapor
Within the wake of the wind farm cluster the air is dryer than in the ambient air outside of the wake (figures 3(b) and (d)). Similar to the wake in the potential temperature, the dryer air is still visible 45 km downwind of the wind farm cluster. Within the wake region the air has a minimum water vapor mixing ratio of 9.8 g kg −1 compared to maximal values of 11.8 g kg −1 outside of the wake. Corresponding to the observations, the model simulates dryer air within the wake region with values around 9.8 g kg −1 . However, the simulations suggest lower water vapor mixing ratios to the west of the wake. The observations show values up to 11.5 g kg −1 whereas the model predicts values in the order of 11 g kg −1 (figure 3(b)).
Associated with the neutrally-stratified layer, dryer air is evident in the vertical cross-section A-B(figures 4(b) and (d)) 5 km downwind of the wind farm cluster. Similar to the potential temperature, it is most likely that the dryer air originated from the dryer layer aloft above 150 mAMSL. This height corresponds to the upper limit of the rotor area, emphasizing that air stemming from above the rotor area is mixed downwards. The mixing of air above the rotor area seems to be be more pronounced in the model than in the observation. Within the upper rotor area, the model simulates a water vapor concentration of 9.5 g kg −1 , whereby the observations show concentrations in the order of 10.2 g kg −1 , indicating that dry air was entrained into too low elevations, due to enhanced vertical mixing into the farms as described in (e.g. Abkar and Porté-Agel 2015, Pan and Archer 2018).

Discussion
This discussion section consists of two parts: the first part discusses the results of the presented case study compared to the other 25 aircraft measurements. The second part estimates the effect of all installed and planned offshore wind farms (i.e. blue and orange wind farms in figure 1) on the MBL for the 10September 2016 case.

Comparison to other cases
Given the results from the case study of the 10 September 2016, one could draw the conclusion that stable conditions are a sufficient constraint to observe a warming and drying at hub height downwind of large offshore wind farms However, this assumption does not hold when analyzing the remaining 25 cases. For example, in the afternoon of the 10 September 2016, the aircraft flew a similar pattern as shown in figure 3, but did not observe any change in temperature and humidity although the atmosphere was stably stratified at hub height ( figure 5(p)). The vertical profiles taken in the morning and in the afternoon differ mainly in terms of wind speed. In the afternoon the wind speed at hub height decreased from 7 m s −1 to values below 6 m s −1 compared to the measurements in the morning, suggesting that the wind speed has to be above a certain threshold to generate enough turbulence to mix the air and induce a warming or drying. Applying these two constraints-stable conditions and wind speeds over 6 m s −1 at hub height to all 26 cases, eleven cases fulfill both criteria. Indeed, in eight of the eleven cases we observe a change in temperature (figure 6). In two of the remaining cases we cannot state for certain that a temperature change did not occur. In the first case (figure 6(f)) we have a strong background gradient in potential temperature hindering the observation of a change in temperature. In the second case ((j), not shown) we have only measurements along the wake and, hence, can not measure any difference between wake and none wake air. In the third case ((k), not shown) we did not observe an impact on temperature, despite the fact that the wind speed was above our defined threshold of 6 m s −1 and the atmosphere was stably stratified at rotor height ( figure 5(k)). However, the measurements were conducted downwind of Godewind (see location in figure 1), a wind farm with fewer wind turbines than the wind farm cluster around Amrumbank West. Consequently, higher wind speeds are necessary to achieve the same amount of mixing. Therefore, we suggest that higher wind speeds would have been necessary to observe a change of temperature at hub height in this case. This assumption is underscored by the observation conducted on the 11April2017 (case (c)), where we measured a warming at Godewind with wind speeds of over 10 m s −1 at hub height.
The height of the inversion (which is partially driven by SST) determines whether the wind farms warm or cool the atmosphere at hub height in stable conditions. In the cases shown in figures 6(a)-(e) a warming is observed at hub height whereby in figures 6(g)-(i) a cooling of the atmosphere was measured. In the warming cases, a pronounced inversion occurred above hub height accompanied by a less stable layer below, indicating that the enhanced negative heat fluxes in the wakes were stronger above than below the The simulations are shown at 10:00UTC corresponding to the start of the observations. Further, the difference in sensible heat flux (e) and potential temperature (f) between a WF and NWF simulation is shown averaged from 08:00UTC to 09:00UTC. In (e) blue colors indicate an enhanced sensible heat flux downward, whereas in (f) blue contours indicate a cooling and red contours a warming caused by the existence of wind farms Black lines in (a), (b) denote the flight track whereas in (e), (f) they show the potential temperature isolines of the WF simulation. The black and thick dashed boxes in (e), (f) show the rotor area of the three wind farms. hub height. As a result, mixing of dry and warm air from above hub height dominates and causes an overall warming at hub height, as schematically indicated in figure 7(a). In contrast, in figures 6(g)-(i) cold SSTs were accompanied by inversions (figures 5(g)-(i)) with their strongest gradients below rotor height, emphasizing that the enhanced negative heat fluxes in the wakes were stronger below than above the hub height, thus causing a net cooling at hub height ( figure 7(b)).
We suggest that these inversions that cause a cooling can be less than 30 m thick. For example, in figure 6(i) we observe a cooling of up to 0.6 K although there is a constant lapse rate within most of the rotor area. However, the SST at FINO1 (see location in figure 1) was 284 K compared to a potential temperature of ≈294 K at 30 m AMSL, indicating that a shallow cold layer close to the ocean surface caused an inversion through the lower portion of the rotor area and below, thus cooling due to the enhanced mixing of this cool air within and partly below the rotor area caused the observed cooling.
The observed warming or cooling is decoupled from the drying downwind. For example, we see a warming in figures 6(a)-(e) and cooling in figure 6(i) but in all measurements we see dryer air downwind, meaning that the moisture flux is decoupled from the heat flux-a result in agreement with the findings of Foreman et al (2017). However, in nine of the cases  that fulfill the criteria for a potential change in temperature at hub height we observed six times a drying and only one time a humidification at hub height. In the remaining two cases we could not measure any change in humidity (case (g), (h)), whereby in case (g) the aircraft was flying at 200m AMSL-a height too high to detect any impact on the humidity ( figure 4(b)).

Future scenario
We have shown that a single wind farm cluster can cause a warming of up to 0.6 K and a drying of ≈0.5 g kg −1 at hub height in the MBL according to the observations. Consequently, the overall effect of all wind farms that are operational, approved or under construction is, hence, of interest. To answer this question, we conducted a simulation for 10 September 2016 including all planned and existing wind farms as they are shown in figure 1 (i.e. orange and blue wind farms). For this case study we have measurements along the cross section A-B (figure 3) and could show that the model simulates the vertical and horizontal impact on temperature and humidity, increasing the confidence in the future scenario. Additionally, we had more warming than cooling cases.
The difference between the WF and NWF simulations (figure 8) suggests a similar warming response (within ± 0.1 K) to the case of 10 September 2016 in the presence of more wind farms. For example, downwind of the large wind farm cluster around Riffgrund, a wide area with a warming of up to 0.5 K is found (figure 8), while downwind of the large westernmost cluster the warming is less than 0.3 K. However, not only the size of a wind farm seems to determine the degree of warming. The wind farm Riffgat with only 30wind turbines causes also a warming of up to 0.5 K.
The simulation suggests a stronger warm air advection aloft at Riffgat and, hence, the inversion at Riffgat is even more pronounced and that in turn allows an even more enhanced downward mixing of warm air. In contrast, the warming at Riffgat is associated with a cooling on the eastern flank of the wake.
While these temperature and moisture changes are novel and may seem consequential when considering the effects of wind farms on local microclimates, it is important to recognize that these are limited effects. The observed local warming and cooling of±0.6 K are small compared to the warming that is caused globally by land cover change (LCC) and land management change (LMC). According to Luyssaert et al (2014) the warming caused by LCC and LMC is in the order of 1.7 K within the planetary boundary layer. Out of the 26flights that occurred over a year, an impact on temperature and humidity only occurred when a strong stable stratification existed at turbine hub height or below rotor height. In other cases, without inversions or with inversions located well above turbine rotor height, the enhanced mixing caused by wind farms would not have such an effect.

Conclusion
This works gives new insights into micrometeorological impacts of large offshore wind farm clusters by the use of aircraft measurements conducted from September 2016 to October 2017 and mesoscale simulations. The main findings include: • Large offshore wind farms can have an impact on the MBL. During five measurement flights, the elevation of the inversion in the rotor disk region was such that the potential temperature increased Figure 8. All existing and all planned wind farms at the North Sea and their micrometeorolocial impact on potential temperature (a) and water vapor mixing ratio (b) they would have had on 10September2016. Shown is the difference between a simulation with wind farms (WF) and a simulation with no wind farms (NWF)-shown is only the region affected by the wind farms. Consequently, warm colors in the potential temperature denote a warming caused by wind farms and gray colors in water vapor mixing ratio indicate a drying of the air introduced by the wind farms.
by up to 0.6 K within the wake of a large offshore wind farm 45 km downwind. This warming was associated with a decrease in the total water vapor mixing ratio by up to 0.5 g kg −1 . In contrast, a shallow inversion below hub height associated with a cold SST causes a cooling of the same magnitude above and at hub height downwind, as observed during three measurement flights.
• These micrometeorological impacts exist only in case of an inversion below or at rotor area. Only in the presence of such inversions can warmer air be mixed downward by the rotors. Depending on the height of the inversions this process is causing either a warming or cooling at hub height. As an inversion acts as a lid for the water vapor evaporating from the ocean, the water vapor concentration is higher underneath the inversion. Consequently, a breakup of the inversion results in a mixing of dryer air downward and, hence, in dryer air within the wake. This process was observed regardless of a warming or cooling, indicating that the moisture flux is decoupled from the heat flux.
• The mesoscale model simulated the observed warming and drying effect of the wind farm cluster reasonably well. Therefore, we could estimate the overall effect of all planned and existing wind farms on the MBL for the 10September2016 case. Even with an increasing number of wind farms, the warming and drying effects remain of the same order of magnitude as in the reference case. The interaction of several wakes resulted in wakes exceeding 100 km in length.
These findings demonstrate that, in some cases, large offshore wind farms can have an impact on the regional microclimate. However, a pure redistribution of moisture and heat has no influence on the regional climate. Only a permanent change in the air-sea interactions could change the regional climate. For example, warmer air over a cold ocean would result in an increased sensible heat flux to the ocean whereby the latent heat flux would transport more water into the atmosphere because of the dryer air within the wake. However, we suggest that these events are rare because a strong inversion at or below hub height is necessary to observe this warming and drying within the wake of large offshore wind farms.