Observations and Simulations of a Wind Farm Modifying a Thunderstorm Outflow Boundary

On June 18, 2019, National Weather Service (NWS) radar reflectivity data indicated the presence of thunderstormgenerated outflow propagating east-southeast near Lubbock, Texas. A section of the outflow boundary encountered a wind farm, and then experienced a notable reduction in ground-relative velocity, suggesting that interactions with the wind farm impacted the outflow boundary progression. We use the Weather Research and Forecasting model and its Wind Farm Parameterization to address the extent to which wind farms can modify the near-surface environment of thunderstorm outflow boundaries. 5 We conduct two simulations of the June 2019 outflow event, one containing the wind farm and one without. We specifically investigate the outflow speed of the section of the boundary that encounters the wind farm and the associated impacts to nearsurface wind speed, moisture, temperature, and changes to precipitation features as the storm and associated outflow pass over the wind farm domain. The NWS radar and nearby West Texas Mesonet surface stations provide observations for validation of the simulations. The presence of the wind farm in the simulation clearly slows the progress of the outflow boundary by 10 over 20 km hr−1, similar to what was observed. Simulated perturbations of surface wind speed, temperature, and moisture associated with outflow passage were delayed by up to 6 minutes when the wind farm was present in the simulation compared to the simulation without the wind farm. However, impacts to precipitation were localized and transient, with no change to total accumulation across the domain. Copyright statement. This work was authored [in part] by the National Renewable Energy Laboratory, operated by Alliance for Sustainable 15 Energy, LLC, for the U.S. Department of Energy (DOE) under Contract No. DE-AC36-08GO28308. Funding provided by the U.S. Department of Energy Office of Energy Efficiency and Renewable Energy Wind Energy Technologies Office. The views expressed in the article do not necessarily represent the views of the DOE or the U.S. Government. The U.S. Government retains and the publisher, by accepting the article for publication, acknowledges that the U.S. Government retains a nonexclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this work, or allow others to do so, for U.S. Government purposes. 20


Introduction
Wind energy deployment is growing rapidly to provide a near-zero emissions source of electricity that can meet increasing energy demands. The International Energy Agency (IEA) predicts wind energy will reach 14% of global capacity (∼1,700 GW) basis for multiple wind farm parameterizations in mesoscale numerical weather prediction models, including the Wind Farm Parameterization (WFP) (Fitch et al., 2012;Fitch, 2016).
The open-source WFP of the Weather Research and Forecasting (WRF) model collectively represents wind turbines in each 60 model grid cell as a momentum sink and a turbulence source within the vertical levels intersecting the turbine rotor disk (Fitch et al., 2012;Fitch, 2016). The virtual wind turbines convert kinetic energy from the wind into power, which is reported as an aggregate sum in each model grid cell. The default setting of the WFP dictates that the turbine-induced turbulence generation is derived from the difference between the power and thrust coefficients, and this option must remain enabled to produce the vertical mixing necessary to attain the expected nocturnal surface warming (Tomaszewski and Lundquist, 2020). Users can adjust the specifications of the parameterized turbine, including its rotor diameter, hub height, thrust coefficients, and power curve, as well as its latitude and longitude location. WFP simulations have been validated with power production data (Lee and Lundquist, 2017a) and airborne measurements of winds (Siedersleben et al., 2018b), temperature and moisture (Siedersleben et al., 2018a), and turbulence (Siedersleben et al., 2020) and have reproduced the observed localized, nighttime, near-surface warming produced by wind turbines mixing warmer air from the nocturnal inversion down to the surface (Fitch et al., 2013; to capture wind farm-near environment interactions makes this model a favorable tool for such a study considering the impacts wind farms may have on outflow boundaries and their resulting changes in temperature, wind, and precipitation.
We hypothesize that wind farms can modify transient and mesoscale features like thunderstorm outflow boundaries. Section 95 2 describes the case study and the model setup. Section 3 presents the modifications to the outflow progression by the wind farm and the impacts to surface temperature, winds, moisture, and precipitation. Section 4 summarizes our results confirming the WRF WFP and radar data capture the wind farm modifying the outflow. The June 18-19, 2019 outflow event near Lubbock, Texas is highlighted in this study as the first known and archived case of an outflow boundary passing over and being modified by a wind farm, which was brought to our attention on social media by Jessie McDonald (@jmeso212). The event began with a cluster of thunderstorms propagating eastward over eastern New Mexico and the western Texas panhandle. These storms formed an organized mesoscale convective system (MCS) around 2300 UTC on June 18 at the New Mexico/Texas border and shifted to move southeastward. An outflow boundary originated from this MCS, 105 visible as a fine line on NEXRAD WSR-88D displays beginning at approximately 2340 UTC (Fig. 1a). This outflow boundary advanced southeastward ahead of the MCS, eventually reaching the Hale wind farm at 0050 UTC on June 19. The wind farm can be detected on the radar display (Fig. 1a) as a cluster of speckled points of high reflectivity, indicative of the hard-target echos of radar beams reflecting off of spinning turbines, known as wind turbine clutter (Isom et al., 2009). A defined notch appeared within the outflow boundary immediately following passage over the wind farm, suggesting a significant reduction 110 in ground-relative velocity where the outflow encountered and interacted with the wind farm (Fig. 1b,c).

Observations Available
The National Weather Service NEXRAD WSR-88D radar in Lubbock, Texas (KLBB), (Klazura and Imy, 1993) provides the initial visualization for this study of the outflow propagating and interacting with the wind farm during the June 2019 event.
Level II radar data (e.g., base reflectivity, base velocity) are provided by the National Oceanic and Atmospheric Administration 115 (NOAA) National Centers for Environmental Information (NOAA National Weather Service, 1991) at 4-minute temporal resolution and quantify the speed and position of the outflow boundary throughout the event.
Surface observations are available through the West Texas Mesonet, a statewide observation network consisting of 40 automated surface meteorological stations that measure up to 15 meteorological parameters over an observation period of 5 minutes (Schroeder et al., 2005). Sampling intervals vary from 3 to 60 s depending on the sensor, and data are reported as 5-minute 120 averages centered on the 5-minute period. A 5-minute observation reporting time has been previously proven sufficient in resolving other density current passages (Toms et al., 2017). The Abernathy surface station is located 5 km southwest of the southwest corner of the wind farm in our study (grey diamond in Fig. 3) and provides 5-minute resolution validation data of 1.5-m temperature, 10-m wind speed and direction, and 1.5-m humidity, among other variables, for the precursor outflow state prior to wind farm interaction for our simulations. We explored accessing meteorological information from the Hale wind farm 125 and others in the vicinity, but those data are proprietary and not available.

Simulations Conducted
We conduct the simulation comprising our study with version 3.8.1 of the Advanced Research WRF (ARW) model (Skamarock and Klemp, 2008;Powers et al., 2017). We define a simulation with three nested domains with horizontal grid spacings of 27, 9, and 3 km, respectively, where the innermost, 3-km domain is centered over the wind farm and outflow event location (Fig.   130 2a). Our previous investigation (Tomaszewski and Lundquist, 2020) of the sensitivity of the WRF WFP to spatial resolution suggests that 3-km horizontal grid spacing is adequate for resolving the wind farm effects. Also based on the results of that study, which argue that the WFP requires fine vertical resolution near the surface, we set the vertical grid spacing to be ∼10 m in the lowest 200 m (Fig. 2b), stretching vertically thereafter, for a total of 58 vertical levels between the surface and 170 hPa. The model time step is 30 s on the outer domain, refined by a factor of 3 for each nest. Turbine-induced turbulence is 135 parameterized via a source of turbulent kinetic energy (TKE). The 0.7 • ERA-Interim (ECMWF, 2009;Dee et al., 2011) data set provides initial and boundary conditions for the simulations, and topographic data are provided at 30-s resolution (nominally 0.8 km at this latitude). Physics options include the Dudhia short-wave radiation (Dudhia, 1989) with a 30-s time step, the Rapid Radiative Transfer Model long-wave radiation scheme (Mlawer et al., 1997), a surface layer scheme that accommodates strong changes in atmospheric stability (Jimenez et al., 2012), the second-order Mellor-Yamada-Nakanishi-Niino planetary 140 boundary layer scheme (Nakanishi and Niino, 2006) without TKE advection, land surface physics with the Noah Land Surface Model (Ek et al., 2003), the single-moment six-class microphysics scheme (Hong and Lim, 2006), and the explicit Kain-Fritsch cumulus parameterization (Kain, 2004) on domains with horizontal grid spacings coarser than 3 km. We simulate the 6-hour window around the time when the outflow passed over the wind farm (June 18 2200 UTC to June 19 0400 UTC). We begin spinup 10 hours prior, at 1200 UTC on June 18.

145
The U.S. Geological Survey Turbine Database (Hoen et al., 2020) provides the latitude-longitude model input locations of the wind turbines at the Hale wind farm (Fig. 2c,d). We use power and thrust curves from the 1.5-MW Pennsylvania State University generic turbine (Schmitz, 2012), based on the General Electric SLE turbine (80-m hub height and 77-m rotor diameter). This turbine model closely matches the 2-MW Vestas turbines actually installed at the Hale wind farm, and Siedersleben et al. (2018b) show little sensitivity to the exact turbine power curve. We assess the impact the wind farm has on   To better understand WRF's skill in simulating the intensity of the outflow event, we plot a time series from the nearby Abernathy West Texas Mesonet surface station against that from the corresponding closest point in the model domain (grey triangle in Fig. 3). Model results are shifted ahead 2 hours to allow for direct comparison between the simulation and observations during the outflow passage, as done in the Arthur et al. (2020) investigation of a frontal passage. The WRF simulation (solid 170 lines in Fig. 4) predicts similar 10-m wind speeds as observed (dotted lines) before the passage in addition to an accurate magnitude of wind speed increase associated with the outflow arrival. The simulated winds remain elevated near ∼23 m s −1 for 15 minutes before decreasing close to the prefrontal state, whereas the surface station observations decrease almost immediately ( Fig. 4a), possibly an artifact of the 5-min sampling in the observations as opposed to the 1-min sampling in the simulation, verified by plotting a 5-minute average of the simulation results (dashed line in Fig. 4a). The simulation displays biases in the 175 2-m temperature and moisture precursor states (Fig. 4b,c). WRF initially has a 2.5 K warm bias, a ∼35% relative humidity (RH) dry bias, and ∼3 g kg −1 dry bias against the observations. These model biases could be due to inaccuracies in the soil moisture that stem from differences in precipitation that occurred earlier in the day. The magnitude of the 2-m temperature decrease ( Fig. 4b) and moisture increases (Fig. 4c) due to the outflow arrival in WRF seem adequate, albeit slightly more intense than in the observations.

Differences in Outflow Passage Between Wind Farm and No Wind Farm Simulations
Having validated WRF's ability to adequately capture the outflow event, we next compare the two WRF simulations to assess the impact a parameterized wind farm has on the simulated outflow. Three instantaneous map views show the difference in 2-m temperature between the simulations, with the no wind farm (NWF) case subtracted from the wind farm parameterization (WFP) case (Fig. 5). Regions of cooler temperatures (blue) indicate that the temperature in the WFP simulation is cooler 185 than in the NWF simulation, suggesting faster movement of the outflow bringing cooler temperatures. Indeed, the dark wind barbs in Fig. 5 representing winds from the WFP simulation indicate stronger winds present (by 5-10 kts) in cooler (blue) regions than in the NWF simulation (light wind barbs). Conversely, red regions indicate warmer temperatures in the windfarm-containing simulation, indicating the outflow is moving slower in this simulation than in the NWF simulation. Early in the outflow event, only subtle differences exist between the simulations upwind from the wind farm (Fig. 5a), likely arising 190 from the generation of gravity waves (Smith, 2009;Meyers, 2018, 2019). These differences increase in magnitude after the outflow passes over the wind farm. A compact region of warmer temperatures (up to 8 K) in the wind farm simulation emerges following outflow passage over the wind farm, indicating that interaction with the wind farm has caused that section of the advancing outflow to slow its speed (Fig. 5b). This region of slowed outflow expands in spatial area as the outflow progresses southeastward (Fig. 5c). A similar speed reduction is visible in the bent outflow shape of the radar observations 195 (Fig. 1b,c). The cooler regions emerging on both sides of the wind during outflow passages suggests flow is being redirected around the wind farm (Fig. 5b). A vertical cross-section of the temperature difference between the simulations taken at 23:22 (dashed line in Fig. 5b) shows that the wind farm (black X) impacts the outflow from the surface to ∼2 km (Fig. 6).
We next sample a point from both the WFP and NWF simulations downwind of the wind farm location (white diamond in Fig. 5) to assess how differences between the simulations evolve at that point following outflow passage (Fig. 7). Close agreement exists between the simulations across all variables plotted preceding arrival of the outflow. Upon the arrival of the outflow, the 10-m wind speed increases first in the NWF simulation (dashed line), reaching a maximum of ∼30 m s −1 . The WFP simulation (solid line) begins its outflow-induced increase a few minutes after the NWF simulation and attains a smaller initial wind speed maximum of ∼25 m s −1 . A secondary pulse of increased wind speeds occurs in both simulation cases and reaches similar magnitudes, suggesting the modified outflow in the wind farm case does not experience notable changes after 205 the initial disruption by the wind farm (Fig. 7a).
The temporal evolution of the 2-m temperature is similar to that of the wind speed. The WFP and NWF simulations produce the same initial temperature until the WFP simulation diverges from the NWF simulation due to the wind-farm-modified outflow approaching ∼3 minutes later. The associated outflow cooling is of similar magnitudes (∼12.5 K) between the simulations, but the WFP simulation reaches its minimum temperature ∼4 minutes after the NWF simulation (Fig. 7b).

210
Differences in the 2-m relative humidity between the simulations evolve similarly as those in the 2-m temperature. Both simulations maintain a value near ∼20% until the passing outflow causes an increase up to 50%, with the increase occurring for the WFP simulation 4-5 minutes after the NWF (green lines in Fig 7c). The absolute moisture quantity (2-m mixing ratio, purple lines) reaches its peak in moisture (10.5 g kg −1 ) ∼6 minutes after the NWF does.
We corroborate the proxies for outflow ground-relative velocity in the time series of meteorological variables (Fig. 5) by 215 directly quantifying the speed of the simulated and observed outflow boundaries (Fig. 8a). We measure the observed outflow speed by tracking the reflectivity fine line along a transect and recording its distance traveled every data update (typically 4 minutes). As the radar is southwest (220°) of the portion of the outflow boundary of interest, which is moving nearly to the southeast at a heading of (120°) (Fig. 1), we note that the feature therefore maintains an approximately constant distance to the radar and thus height above ground even as the feature moves, thus not impacting our calculations of ground-relative outflow 220 speed.
Without a fine line present in the simulations to denote the outflow boundary, we choose to track the simulated outflow using the spatial gradient in wind speed, specifically the 4 m s −1 km contour (e.g., Fig. 8b). The simulations are examined at 4-minute intervals to match the temporal resolution of the radar data. Both simulation and radar outflow are measured against a 5-kmby-5-km grid to estimate distance traveled (see Fig. 8b). The transect along which we measure distance traveled is oriented to averages of each case are plotted in Fig. 8a, around which ±1 standard deviation forms the shaded cloud and serves as our error bounds. As in Fig. 4 and Arthur et al. (2020), the simulation results are shifted forward 2 hours to align with the radar results. A running average with an 8-minute window was applied to all three time series to smooth the results for viewing.
As suggested in Fig. 5 and Fig. 7, speeds of both simulation cases and the radar data begin at similar values near 80 km hr −1 (Fig. 8a). The simulated and observed outflows decelerate slightly as they propagate away from the source thunderstorm. When 235 the radar outflow (blue line) encounters the wind farm, its speed reduces from 60 km hr −1 to nearly 40 km hr −1 . The radar outflow recovers within 10 minutes back to >60 km hr −1 before being obscured by precipitation. Similarly, the Wind Farm Parameterization (WFP) simulation (orange line) fluctuates around 70 km hr −1 until encountering the wind farm, when it then drops in speed to about 40 km hr −1 . The WFP simulation experiences a larger reduction in speed than observed but reaches its speed minimum ∼8 minutes later than the observations. Additionally, the WFP simulation recovers its speed twice as slowly 240 as the observed outflow. Such delays in the WFP outflow evolution could be artifacts of the 3-km model grid spacing or more likely the initial and boundary conditions. The no wind farm (NWF) simulation (green line), lacking wind farm interference, maintains a ground-relative velocity between 60 and 75 km hr −1 throughout the period of interest.

Simulated Impacts of Modified Outflow Boundary to Precipitation
Subtle but significant impacts of wind-farm-modified outflow to meteorological variables like wind speed, temperature, and 245 moisture outlined in Sec. 3.2 prompt the question of the extent to which a wind-farm-modified outflow boundary can impact precipitation location and quantity. We address this question by integrating the total precipitation over a 100-km radius around the wind farm and comparing these quantities for the WFP (green line) and NWF (black dashed line) simulations every minute ( Fig. 9a) and accumulated in time (Fig. 9b) over 3 hours. While the 1-minute precipitation totals across the region differ slightly between the simulations, the total accumulated precipitation remains unchanged despite the altered outflow in the WFP case. 250 We conclude that the introduction of roughness elements may change the distribution of the precipitation by a maximum of ∼1 cm across the domain at a single moment in time (Fig. 9a), but the overall precipitation accumulation is unaffected (Fig. 9b).
Furthermore, a histogram detailing the number of 3-km grid cells that do experience a change in precipitation at a 1-minute moment in time over 3 hours due to the presence of the wind farm reveals that no single grid cell experiences a delta greater than ±7 mm, and over 93.4% of grid cells experience 0 change in precipitation (Fig 10). Changes to precipitation due to the 255 wind farm are thus both transient and localized.

Power Production at the Simulated Wind Farm
Given that wind farms can modify outflow and their associated meteorology, we next explore the effects an incoming outflow can have on a wind farm and its power production. Time series of the simulated 80-m wind speeds from all turbine-containing grid cells in the Wind Farm Parameterization (WFP) simulation indicate that several grid cells exceed the wind turbine's cut-out 260 speed (25 m s −1 , black dashed line in Fig. 11a), most notably at 23:10 and 23:35 UTC. Winds in excess of this cut-out speed force the turbines to brake their blades to prevent structural damage, halting power generation. The corresponding time series of power from turbine-containing grid cells and the total integrated farm power (Fig. 11b) reflect this reduction in power during those times when the cut-out wind speed is reached. Power data from the Hale wind farm are proprietary and unavailable for validation, though simulation data suggest outflow winds are high enough to cause wind turbines to cut out and reduce total 265 farm power generation (Fig. 11). simulations immediately following outflow boundary passage over the wind farm, corresponding to the dashed line in (Fig. 5b). Note the y-axis ticks are not spaced linearly due to the increasingly coarse vertical grid spacing at higher model levels.

Discussion and Conclusions
Increasing deployment of wind energy necessitates obtaining further knowledge on the environmental impacts of wind farms to ensure their long-term sustainability and suitability. A lower-atmospheric phenomenon not yet explored in relation to interacting with wind energy is thunderstorm outflow. Herein, we assess the impact a wind farm can have on outflow movement via 270 observations and simulations.
We first observed wind farm impacts to outflow in NEXRAD WSR-88D radar reflectivity. On June 18, 2019, a section of an advancing outflow boundary visible on radar encountered the Hale wind farm near Lubbock, Texas and decelerated in response.
We ran two Weather Research and Forecasting (WRF) simulations to capture this event: one with a Wind Farm Parameterization (WFP) enabled and another with no wind farm present (NWF). Using observations from a West Texas Mesonet surface station, 275 we verified that the simulations were producing reasonable solutions of the outflow event and could be used to quantify the extent a wind farm can modify propagating outflow.
Just as with the radar reflectivity, spatial differences between the WFP and NWF simulations exhibited a similar pattern, indicating the wind farm slowed the progress of the outflow boundary (Fig. 1, Fig. 5). Time series of simulated surface wind speed, temperature, and moisture revealed that perturbations associated with outflow passage were delayed by up to 6 minutes 280 when the wind farm was present in the simulation (Fig 7.)  after encountering the wind farm, whereas the NWF simulation maintained near constant speed throughout the period (Fig.   8). Impacts to precipitation were minimal, with no change to total accumulation across the domain (Fig 9). Localized shifts to precipitation location in the WFP simulation caused a maximum instantaneous grid-cell precipitation difference of 7 mm 285 km −2 , but 93.4% of grid cells within the area over the event period experienced no change in precipitation (Fig. 10).
While we have shown that a wind farm can interact with and modify thunderstorm outflow, impacts to the modified outflow speed and associated kinematic and thermodynamic variables are transient and localized. These subtle changes arising from wind farm interaction may be useful to consider when conducting nowcasting of precipitation and wind speed on a scale of a few kilometers and minutes, perhaps for aviation or other time-sensitive purposes. Impacts beyond that scale appear to be This study uses a single known case of a wind farm interacting with outflow and is corroborated by simulations of that case. This case study could motivate a larger-scale climatology of additional outflow-wind-farm interaction events, including different environments with variable soil moistures or other meteorological properties. Such a climatology could consider wind farms of different layouts and sizes, as well as different turbine types and sizes to assess generalized sensitivity of atmospheric 295 modifications to the turbine layouts and density. In previous studies (Lundquist et al., 2018), we have noticed wind farms apparently modifying the passage of frontal boundaries, so a large-scale climatology of such events, tracking frontal groundrelative velocity, could shed more light on how widespread and impactful the modification of atmospheric processes by wind farms can be.
Code and data availability.