Validation of Doppler Wind Lidar Measurements with an Uncrewed Aircraft System (UAS) in the Daytime Atmospheric Boundary Layer
1. Introduction
Doppler wind lidar (DWL) systems are used in a broad spectrum of applications like quasi-operational measurements of the wind vector at meteorological observation sites (Päschke et al. 2015), wind profiling from satellites (Chanin et al. 1992; Witschas et al. 2020), detection of gravity waves in the strato- and mesosphere (Baumgarten et al. 2015), monitoring wind shear for flight safety (Nechaj et al. 2019), or wind energy applications (Käsler et al. 2010; Liu et al. 2019; Wildmann et al. 2019). Numerical weather prediction (NWP) models need vertical profiles of state and process variables in the atmospheric boundary layer (ABL) to improve the prescription of initial conditions for the simulations (Knist et al. 2018).
Common measuring strategies for deriving the horizontal wind vector from DWL measurements are the Doppler beam swinging (DBS) and the velocity–azimuth display (VAD) method (Reitebuch and Hardesty 2021). Both scanning strategies rely on the assumption of homogeneity and stationarity of the horizontal wind vector in the analyzed air volume during the scan. However, these conditions are usually not fulfilled in reality, especially over heterogeneous terrain. Hence, a detailed quantification of the statistical uncertainty and systematic errors of DWL data is required for the different scan strategies and atmospheric conditions (Rahlves et al. 2022). Since calibration and validation of remote sensing instruments in a laboratory are impossible due to their range, resolution, and volume-sensing approach (Reitebuch and Hardesty 2021), there is a need for other validation methods.
There are various approaches to validate DWL data both with numerical simulation and in a real-world scenario. Large-eddy simulations (LES) have been extensively employed to identify DWL errors and uncertainties for different applications and specific flow conditions by implementing virtual lidar models in the simulation domain (Lundquist et al. 2015; Gasch et al. 2020). Rahlves et al. (2022) used LES and showed that the statistical uncertainty of DWL data decreases with increasing averaging time. Moreover, an analysis taking into account the wind field variability highlighted a statistical error in the O(1) m s−1 for 10-min averaged wind speed in a moderately convective ABL, while exceeding 2 m s−1 in strongly convective conditions. Cohn and Goodrich (2002) compared radar wind profiler (RWP) and DWL data showing a standard deviation of 0.88 m s−1 between both systems for the horizontal wind estimation. However, this result is based on a single 2.3-h sample and did not account for variations caused by atmospheric conditions. Päschke et al. (2015) compared a 1-yr dataset of 30-min averages of DWL data with data from an operational 482-MHz RWP and revealed a good agreement between the two systems, with root-mean-square differences ranging between 0.5 and 0.7 m s−1 for wind speed and between 5° and 10° for wind direction. Due to the lower altitude limit of the RWP, no comparisons were available for altitudes below 500 m above ground level (AGL) and shorter temporal averages.
Another possible validation method is through the usage of radio soundings. In a 2-day experiment, Ishii et al. (2005) showed a good agreement between DWL and radiosonde data, with a mean bias of 0.2 m s−1 and 1.4° and a standard deviation of 1.4 m s−1 and 3.5°. Mariani et al. (2020) compared in a 19-month campaign, two DWL scanning methods with radiosonde data at three locations in the Arctic region. In their study, both scanning methods utilized 70° elevation angle: the VAD method (eight 45° spaced beams) exhibited a smaller mean bias (−0.27 m s−1) than the DBS method (−0.46 m s−1), both compared to the radiosonde. In the Lidar Measurement Campaign Sola (LIMECS), Kumer et al. (2014) found a strong correlation of horizontal wind between altitudes of 500 and 1325 m, but a weak correlation (correlation coefficient R = 0.6) below 500 m. Mariani et al. (2020) and Kumer et al. (2014) observed that the comparability of horizontal wind measurements below 500 m is challenging due to the radiosonde’s pendulum motion. Moreover, the vertical resolution is coarse; there is no temporal averaging; and the radiosonde’s drift increases the horizontal displacement, collectively contributing to biases. Comparison studies using cup anemometers are limited to the ground (Lawrence et al. 1972), or to the maximum height of a meteorological tower such as 100 m in Smith et al. (2006), or 116 m in Gottschall et al. (2012), Courtney et al. (2008).
Ultimately, there is a gap in data availability at an altitude range between 100 m (towers) and 500 m (due to the unavailability of RWP data and coarse, biased radiosonde data). This gap includes the influence of atmospheric stability on DWL systems, which, according to virtual lidar models employed in LES simulations, is not negligible (Robey and Lundquist 2022) but has not been verified with experimental data (Klaas-Witt and Emeis 2022; Gasch et al. 2020). An accurate validation within this altitude gap and across different ABL stability conditions can be achieved using high-resolution in situ airborne wind measurements by uncrewed aircraft systems (UASs). As pointed out in Bange et al. (2021), a small research UAS delivers as accurate thermodynamic and turbulence data as a large crewed research aircraft although carrying a smaller scientific payload and endurance but at much lower operational costs and logistic efforts. Furthermore, a small research UAS is capable of measuring the three-dimensional turbulent wind vector with very high temporal and spatial resolutions without the disturbing effect of a large fuselage, wings, and propellers (Platis et al. 2016; Rautenberg et al. 2019; zum Berge et al. 2021; Schön et al. 2022). Small UASs were already applied in the past for remote sensing validation (Martin et al. 2011). An overview of meteorological UAS and research aircraft, in general, can be found in Bange et al. (2013, 2021).
In this study, in situ turbulent wind vector measurements were obtained using a UAS of type Multi-Purpose Airborne Sensor Carrier-3 (MASC-3) (Rautenberg et al. 2019). These measurements were carried out by the Environmental Physics Team of the University of Tübingen, Germany, during three intensive observation periods (IOPs) in July 2020, June 2021, and November 2022, in the frame of the Validation of numerical simulations and remote sensing using UAS (VALUAS) project. The goal was to use data provided by the UAS as a reference for assessing the wind speed and direction accuracy between 100 and 500 m AGL of two DWLs operated by the German Weather Service [Deutscher Wetterdienst (DWD)].
The experiments’ motive was for the following questions to be addressed:
-
How much does the spatial separation of the DWL beam and the UAS flight path limit the comparability?
-
Is there an altitude dependence of the intercomparison results between the DWL and UAS wind measurements?
-
Does atmospheric stability influence the intercomparison results?
The manuscript is structured as follows: in section 2, the instruments (UAS, meteorological tower, and DWL including scan strategy), the site, and the data analysis methods are presented. Section 3 shows and discusses the results of the VALUAS IOPs including a quality control of the UAS wind measurements using the meteorological tower instruments at Falkenberg. The overall performance of two DWLs is analyzed and compared to UAS in situ data taking into account several variables such as the UAS fetch angle, flight altitude, and atmospheric stability. In section 4, the conclusions and recommendations for future studies are given.
2. Materials and methods
a. Measurement site and experimental setup
All meteorological measurements described in this study were conducted at the boundary layer field site Falkenberg (in German: Grenzschichtmessfeld, GM Falkenberg) of the DWD, about 60 km southeast of Berlin, Germany (Fig. 1). Flat farmland with few forest patches and small villages in between dominates the landscape around the measurement site (Mengelkamp et al. 2006). The DWD operates this site to investigate the ABL in the rural landscape around Lindenberg to complete the so-called “Lindenberg Column” reference profile measurements of the whole troposphere (Neisser et al. 2002). Three IOPs at the GM Falkenberg were conducted in the framework of the project VALUAS. The first campaign (IOP-1) took place from 7 to 29 July 2020, the second (IOP-2) from 7 June until 2 July 2021, and the third (IOP-3) from 17 to 29 November 2022 (list of flights available in appendix A). During these IOPs, two DWLs of type Halo Photonics Streamline were deployed next to the 99-m meteorological tower (Fig. 2). The free space area of the site was also used for the UAS of type MASC-3 takeoff and landing procedures to conduct the necessary measurement flights for comparison.
The position of the measurement site, GM Falkenberg, about 60 km southeast of Berlin, Germany.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0127.1
Top of the meteorological mast at the MOL-Richard-Aßmann Observatory (RAO) site at GM Falkenberg, close to Lindenberg, with the installed cup anemometers and wind vane at 98 m above the ground. Picture was taken with a DJI mavic the morning of 25 Nov 2022.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0127.1
b. Meteorological tower
The GM Falkenberg hosts a 99-m high meteorological tower at 52°10′N, 14°07′E, 73 m above mean sea level (Fig. 2). The tower provides horizontal wind measurements with cup anemometers (Thies 4.3303.22.000) at 10, 20, 40, 60, 80, and 98 m and wind vanes (Thies 4.3121.32.000) at 40 and 98 m. Water vapor and air temperature are measured using Vaisala HUMICAP humidity and temperature probes of type HMP45 in combination with R. M. Young model 43408 aspirated radiation shields at 10, 20, 40, 60, 80, and 98 m. Surface pressure at the tower (1 m above ground level) is measured by a Vaisala PTB220A barometer. Data of wind speed and wind direction were available as 1- and 10-min averages, while 10-min averages were used for the other atmospheric variables.
c. Doppler wind lidars
For this study, DWD provided data from two Halo Photonics DWL systems: a regular Streamline unit (S/N 177) owned by DWD and a Streamline XR unit (S/N 143) as a contribution from the Technical University of Berlin to the Field Experiment on Submesoscale Spatio-Temporal Variability in Lindenberg (FESSTVaL) campaign. Therefore, WL143 was present for the 2021 campaign only, while WL177 was operated during all three IOPs.
To derive the mean wind using DWLs, the Meteorological Observatory Lindenberg (MOL) uses a best practice VAD variant, which has proven to be particularly efficient over years of testing. The variant used in Lindenberg is described in Päschke et al. (2015) and Teschke and Lehmann (2017). This method allows for a simple calculation of uncertainties as well as efficient quality control via common fit parameters, namely, the coefficient of determination (R2) and the condition number of the algebraic least squares regression (CN). In addition, this VAD utilizes the minimum number of directions available for the calculation as a consistency criterion. For example, with 24 viewing directions and the minimum number of directions set to 12, there are practically no outliers in the wind retrieval.
WL143 uses DWD’s VAD variant with 75° elevation and 24 different scan directions, which takes around 120 s to complete a full azimuth circle, while WL177 applies a continuous scanning mode (CSM) at 62° elevation designed for measuring wind gusts, which completes a full azimuth circle in just 3.4 s. Due to their different pulse durations, system type, and scan mode, both lidar products differ in their height resolution and maximum height. Additionally, the lowest gate in each lidar makes up the blind zone. Hence, usable data start at 70 and 40 m for WL143 and WL177, respectively (see Table 1). It is important to note that due to their different configurations, a single line-of-sight (LOS) measurement of WL143 averages 30 000 individual pulses, while WL177’s rapid scanning uses only 3000 pulses. This has influence on the measurement quality and requires a different filter method prior to the wind retrieval because filtering data with the common signal-to-noise ratio (SNR) approach would result in a marked loss of data availability. Therefore, the retrieval for both lidars was conducted with the consensus filter method. This filter is based on the random sample consensus (Fischler and Bolles 1981), and it is used operationally in DWD’s radar wind profiler network. The wind processing was conducted with DWD’s Doppler lidar processing toolbox11 using a time window of 10 min. The toolbox incorporates the consensus filter and can process conical scans like VAD and CSM.
Measurement properties of the DWL instruments WL177 and WL143 deployed at GM Falkenberg (WL177 during all IOPs, WL143 during IOP-2 only).
d. UAS platform
For obtaining the in situ horizontal wind velocity, the fixed-wing UAS platform MASC-3 (Wildmann et al. 2017; Rautenberg et al. 2019; zum Berge et al. 2021; Schön et al. 2022; zum Berge et al. 2023) was employed (Fig. 3). The scientific payload consists of a five-hole probe (5HP), an inertial measurement unit (IMU), and a fast thin-wire resistance thermometer operated together to get high-resolution wind measurements (van den Kroonenberg et al. 2008; Wildmann et al. 2014a,b). MASC-3 uses a sampling frequency of 100 Hz to measure the three-dimensional wind vector and air temperature. The three-dimensional wind vector in Earth’s coordinate system is calculated using raw data from the 5HP, IMU, and GPS data, along the three-dimensional flight path of the UAS. Rautenberg et al. (2019) presented a detailed description of the complete MASC-3, its sensor system, and the raw data postprocessing procedures.
The UAS platform MASC-3 in flight (2-m length, 4-m wingspan, 8-kg takeoff mass, and 18.5 m s−1 constant true airspeed). The standard turbulence instrumentation is mounted at the very front of the aircraft. The pusher–engine configuration with two rotor blades is visible at the tail of the UAS. Courtesy of Ines Schäfer.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0127.1
Flight strategy
Measurement flights were carried out by repeatedly flying along straight and level ground tracks (legs) at various altitudes between 89 and 550 m above the ground (Fig. 4), with a constant indicated airspeed of 18.5 m s−1. Each flight leg has a duration of at least 1 min. To compare flight leg data to the averages provided by the cup anemometers at the tower and to the DWLs, several consecutive flight legs at the same altitude were combined to cover a measurement time of at least 10 min.
The three-dimensional extent of all the UAS legs used for this study. The red column represents the meteorological 99-m tower.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0127.1
e. UAS data processing
All flight legs were initially sorted into different altitude groups. Since this study aimed to validate DWLs with UAS measurements, the groups’ vertical extent was set to be as close as possible to a multiple of the height resolution of the WL177 (the one with the most extensive dataset). The extent was set accordingly to approximately 50 m.
The flight legs within a specific altitude group were then divided into suitable time intervals. The length of these intervals varies based on the desired averaging time. Executing this process was complex as legs at the same altitude tended to have varying duration times. This occurred because the UAS maintained a constant indicated airspeed, causing the ground speed and, consequently, the duration of each leg to differ depending on factors like wind direction, flight direction, and wind speed.
Finally, for every time interval and altitude group, the median of wind speed and wind direction was calculated from the UAS dataset. The results of the calculation were then compared with tower and DWL data for the same altitude range and time.
1) Averaging times
As explained in the previous section, each and every flight leg has a different duration due to the flight conditions. The procedure to get the leg bins varies, depending on the averaging time.
Legwise average: Considering the highest resolution of cup anemometer data (1 min), a leg is considered valid if it lasts at least 60 s, ensuring a unique identification for comparison with the anemometer dataset. However, it is even possible for single legs to exceed 300 s in duration due to encountering strong headwinds. In such cases, the decision was made to avoid dividing the legs into multiple subsections. This choice is completely arbitrary. One can also decide to divide the legs into units of 60 s, but then other problems arise, such as if a leg is not an exact multiple of a minute, which part of the leg should be discarded? Instead, an easier and more convenient approach involved averaging the cup anemometer data collected throughout the entire leg. The distribution and mean value of the legwise average at 98 m AGL for UAS quality assurance are presented in Fig. 5a.
Distribution (blue histogram) and mean value (vertical black dashed line) of the leg bin duration for the three different time averages at 98 m AGL.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0127.1
The 10- and 30-min averages: Given that there are no individual legs that last exactly 10 or 30 min, our approach involves considering consecutive legs and creating a representative interval of either 10 or 30 min. To accomplish this, we opted to group together the number of legs that, when combined, come closest to the desired average duration. This strategy may sometimes result in a group duration that is lower or higher than the required average, but it would still be closer to the 10- or 30-min average compared to adding or removing another leg to the group. Figures 5b and 5c display histograms depicting the distribution and mean value of leg groups for the UAS quality assurance at 98 m AGL with 10- and 30-min averages. These histograms demonstrate that while shorter leg groups are present, the overall average duration aligns with the respective requested time average.
For the quality assurance of the UAS measurements (comparison with tower data), all three different averaging periods were used: legwise, 10 min, and 30 min. For the DWL validation, the investigation of the crosswind, altitude, and atmospheric stability effects, 10-min data averages from UAS and DWL were used. Table 2 gives an overview of the different systems and the availability of data products for the three averaging periods, while appendix B shows a schematic flowchart of the UAS data processing procedure.
Averaging times and their usage in this study for each system.
From the UAS leg groups, the first timestamp of the first leg in the group and the last timestamp of the last leg in the group were used as the start and end time. Then, depending on the time average, the data from the other systems were either picked up as a single value (DWLs and 10-min tower data) or averaged if more values were present between the start and end time (1-min tower data).
2) Spatial separation effect (fetch angle)
When comparing stationary measurements of a cup anemometer or DWL with data from a moving platform like a UAS, the effect of spatial separation has to be considered. The stationary sensor probes the air transported by the mean wind vector across its position (advection). On the contrary, the UAS is moving through the air quickly enough so that the gathered data might be interpreted as a snapshot of the air mass. It depends on wind direction, speed, and the relative position of the UAS flight path with respect to the stationary instrument, whether the probed air masses are identical or not.
The flight paths for this study were designed in a way that the UAS was always flying toward or away from the location of the stationary measurements (tower or DWL) on repetitive flight legs, ideally aligning the ground trajectory with the mean wind direction. Aircraft controlled by an autopilot system can follow a preprogrammed ground track with high accuracy, independently of the current wind direction in the atmosphere. Hence, despite a careful mission planning, there could be an angle α (fetch angle) between the flight direction (track) of the UAS and the wind direction ϕ.
This angle α can range from 0° (headwind or tailwind) to 90° (crosswind) on repetitive legs. Headwind is coming from the opposite direction the aircraft is flying, parallel to its flight direction. Tailwind is going in the same direction as the aircraft, parallel to its flight direction. Pure crosswind crosses the aircraft track perpendicularly without any headwind or tailwind components.
The aircraft moves through the air, while the air itself moves in Earth’s coordinate system due to the mean wind vector. In the case of α ≈ 0 (head or tailwind), the aircraft is probing exactly the same air mass that is advected over the stationary measurement sensor. Part of the air moves perpendicular to the aircraft’s ground track at higher crosswind angles. Then, the aircraft probes different air masses on several repetitive legs compared to the one advected over the stationary measurement sensor.
Besides that, the probed air continuously exchanges energy and momentum with the surrounding air (depending on the local turbulent mixing) and the heterogeneous surface. The degree of similarity of influence onto the measured airborne data and the stationary instruments differs with the fetch angle α. For α ≈ 90°, the surface to the left or right of the ground track influences the measurements about as strong as the surface along the ground track.
Comparing aircraft measurements with sensors that are stationary at the ground (e.g., towers or DWL) requires considering that for small fetch angles α, the air is advected along the aircraft track toward the stationary sensor, in case the aircraft is moving toward or away from the tower or DWL. More prominent fetch angles mean that the air advected toward the stationary sensor only crosses the aircraft’s track at one location. This raises the question of whether the agreement of the UAS and tower or DWL data depends on the fetch angle α between the wind direction and flight direction.
Therefore, before any comparison with any stationary sensor, an analysis of the influence of the fetch angle was performed.
f. Statistical measures
To compare datasets from the tower, UAS, and DWL measurements on a statistical basis, suitable statistical quantities (measures) are defined in the following, both for horizontal wind speed υ and direction ϕ.
1) Wind speed
The meteorological wind vector is defined by u = (u, υ, w) with u being positive eastward, υ being positive northward, and w being positive upward. However, since the vertical wind component is not the subject of the present study, mainly the horizontal wind vector is analyzed and referred to in polar coordinates using the wind speed variable υ = |(u, υ)| and direction ϕ (with ϕ = 0 for wind coming from the north).
2) Wind direction
g. Atmospheric stability measures
The lapse rate γ and the bulk Richardson number Rib are used to investigate the influence of atmospheric stability on the DWL measurements. The first measures static stability due to temperature gradients over a specific height difference (thermal stratification). In contrast, the second one also considers the horizontal wind speed gradient (shear) in the vertical direction (dynamic stratification).
3. Results and discussion
a. Quality assurance of UAS data
Before UAS data are used to validate DWL data, the UAS measurements were compared to tower data for quality assurance.
Quality control was performed by comparing the horizontal wind speed and direction measured by UAS with the tower measurement at 98 m AGL. Time-averaged cup anemometer data were compared to the corresponding flight legs at 98 m during the same time. The averaging periods were defined as the duration of one flight leg (legwise, see above), 10, and 30 min.
1) Fetch angle analysis
At first, an analysis of the spatial separation effect was performed on the legwise and 10-min average data (the 30-min average was not used due to the small size of this dataset). Results are reported in Table 3. This analysis shows that the RMSD for both wind speed and direction increases with the fetch angle. The agreement of the wind speed measurements is more affected than the wind direction measurements.
Statistical measures evaluating the fetch angle α effect for the UAS quality assessment. The α limits are chosen in such a way to split as evenly as possible the data pairs n between the different groups.
To have a good compromise between the data quality and the dataset’s size, it was chosen to use only the data with a fetch angle α lower than 39.3° for the further analysis.
2) Final quality assessment
Figures 6 and 7 show the result of the comparison for both wind speed and wind direction for the filtered dataset (fetch angle α lower than 39.3°): each data point in the diagrams represents one pair of UAS and tower measurements. Table 4 presents the corresponding statistical measures.
Comparison of UAS wind speed data with cup anemometer data at 98 m AGL. (a)–(c) Different averaging intervals explained in section 2e(1). The red lines are the result of linear regressions (same in the following diagrams).
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0127.1
Comparison of UAS wind direction data with vane data at 98 m AGL. (a)–(c) Different averaging intervals explained in section 2e(1).
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0127.1
Statistical measures describing the quality of the UAS as a horizontal wind vector sensor. For the wind direction, only data where the wind speed is higher than 2 m s−1 are considered.
The normalized mean bias BY for wind speed is always less than 4%. With a mean wind speed of 6 m s−1, this equals to an error of 0.24 m s−1. Such a small value indicates that any systematic difference between UAS and tower measurements regarding wind speed is negligible. The normalized RMSD σY is about 20% for the legwise average. This high initial value is probably caused by atmospheric turbulence. The correlation between the tower-mounted cup anemometer and the UAS-based 5HP measurements is always more than 0.9. When increasing the averaging period to 10 and 30 min, the correlation coefficient Rυ grows and σY drops: the statistical agreement between both systems increases strongly, mainly due to the low-pass filtering of turbulent motion. The fact that the 30-min case has the highest correlation, despite the low number of data points, implies that a lower correlation in the shorter averaging times is mainly due to the random turbulence error in the convective boundary layer, which affects measurements due to spatial separation.
The wind direction also shows a high degree of agreement between tower and flight data, as long as data points with wind speeds below 2 m s−1 (black triangles in Fig. 7) are removed when calculating the statistics. Low wind speeds naturally have a high variation in the wind direction, thus making comparability unfeasible. For the wind direction, the correlation reaches already 0.98 using legwise averaging. The root-mean-square difference σϕ is about 17.5° due to turbulent fluctuations, dropping to 9.9° for a 30-min averaging. The mean bias Bϕ drops from 4.7° for the legwise averages down to 2.7° for a 30-min averaging, indicating that there is very low significant systematic difference between tower and UAS-measured wind directions.
The quality assurance of the UAS measurements using the tower data for comparison highlights that despite the different measurement principles, the UAS is a well-suited tool for DWL validation.
Considering that the next comparison involves 10-min average data, our reference system can be characterized by negligible BY with σY = 13.7% and Bϕ = 3.3° with σϕ = 13.5°.
b. Doppler wind lidar validation
As described in section 2e(1), consecutive UAS flight legs at a constant altitude were grouped to calculate the 10-min averaged data for the DWL validation, as the operational DWL dataset provided by the DWD consists of 10-min averages. This procedure leads to a database consisting of averaged data pairs (UAS, DWL) of wind speed and direction for both types of DWL measurements.
In the first step, all available altitudes from the whole dataset are compared. 53 (UAS, WL143) and 275 (UAS, WL177) data pairs are available for the wind speed comparison. This difference is because WL143 was not operated during the IOP-1 and IOP-3 in 2020 and 2022.
In the case of wind direction, for WL177, only data from 2021 to 2022 are shown and analyzed since, in 2020, technical issues with wind-direction derivation for this device occurred. This technical problem, combined with the criterion of taking pairs where the wind speed is above 2 m s−1, reduced the number of wind direction data pairs (UAS, WL177) to 182 and (UAS, WL143) to 41.
1) Fetch angle analysis
To evaluate the effect of spatial separation, the database consisting of data pairs of UAS and DWL measurements averaged over 10 min was split into four parts, similarly to section 3a(1). Since the data availability of the WL143 is not as good as the one of the WL177, only the WL177 system is taken into account here. The effect of the spatial separation due to the fetch angle α is expected to be similar for both DWL systems since they are situated next to each other at the measurement site.
Table 5 lists the result and statistical measures for the WL177. The statistical measures R, B, σ show an increase of σY and σϕ with the rising fetch angle. The correlation coefficient Rυ for the wind speed decreases, down to 0.83 for fetch angles from ≈40° to 90°, while BY does not show a clear tendency. In contrast to this, the correlation coefficient of the wind direction shows a slight decrease only for the fourth group (≈40° to 90°), which contains the most significant fetch angles α, i.e., strong crosswind flights. The bias of the wind direction between WL177 and UAS also does not show a clear tendency.
Statistical measures evaluating the fetch angle α effect on the wind speed and direction data for the UAS vs WL177 comparison. The α limits are chosen in such a way to split as evenly as possible the data pairs n between the different groups.
Summing up, the measures R, B, σ that describe statistical and systematic differences between the UAS and the DWL data are affected by the fetch angle α (or crosswind flights), but as long as α < 39.3°, the influence is considered acceptable. This condition reduces the usable wind speed datasets from 275 to 205 (UAS, WL177) and from 53 to 46 (UAS, WL143), while for wind direction, it is from 182 to 144 (UAS, WL177) and from 41 to 38 (UAS, WL143).
2) Overall validation
Figure 8 shows scatter diagrams for wind speed and direction for both DWL instruments compared with UAS measurements. Table 6 lists the corresponding statistical measures.
UAS wind speed and direction data compared to WL143 and WL177 data, respectively, for 10-min averaging intervals. See Table 6 for the corresponding statistical measures.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0127.1
The statistical measures of the overall comparison of UAS with both DWL systems. These measures statistically describe the scatter diagrams of Fig. 8.
For both DWL, the correlation with the UAS data is high: more than 0.94 for the wind speed and more than 0.98 for the wind direction and comparable with the agreement between tower and UAS data.
The WL143 statistical scatter for wind speed σY is more significant than the WL177. In contrast, the scatter for wind direction (σϕ) is smaller. The reason might be the different scanning strategy (see section 2c) or the different turbulent conditions since the two datasets do not cover identical meteorological situations.
The systematic differences between UAS and DWL in wind speed are almost one order of magnitude larger than the UAS–tower agreement, especially for the 10-min average. According to Eq. (1), the WL143 shows a 5.1% tendency to overestimate the horizontal wind velocity. At the same time, the WL177 tends to underestimate it by 5.7%. The 5.1% underestimation in WL177 can be attributed to a noise-induced bias that becomes increasingly noticeable in the case of weak signals (Päschke and Detring 2023, manuscript submitted to Atmos. Meas. Tech. Discuss.). For WL134, the amount of data pairs for the comparison is only a fifth of the number of the data pairs available for WL177. Additionally, no wind speed was much greater than 10 m s−1. Therefore, the point cloud is more compact and the linear regression is not as stable as for the comparison with WL177 and hence, the results are less reliable. This hints that for a reliable comparison, the data points should cover a considerable range of velocities. The systematic differences between the reference platform and the DWL wind direction are almost negligible, even though the WL177 seems to be the more accurate.
3) Simultaneous validation
The dataset was cut-filtered to investigate how both systems compare with the same atmospheric conditions by considering only situations where the two lidars operated simultaneously (viz., only during IOP-2 in 2021). This filtering does not reduce the size of the (UAS, WL143) dataset but discards most of the data pairs (UAS, WL177). The final number of entries is 45 for the wind speed and 38 for the wind direction.
The statistical measures of the simultaneous comparison are shown in Table 7. The statistics are consistent with Table 6 and confirm the conclusions drawn in section 3b(2).
The statistical measures of the simultaneous comparison of UAS with both DWL systems.
It should be noted that WL143 operates with a smaller zenith angle: this implies that the projection of the radial winds on the horizontal wind vector is smaller and thus probably more uncertain, especially in the convective ABL. Moreover, a 10-min average for WL143 just contains five scan cycles against more than 170 for the WL177.
c. Effect of the altitude
In this section, the 10-min averaged data pairs are used to investigate whether a dependency of the overall data quality on the scanning altitude is present. The data pairs were subdivided into altitude groups and compared to the closest DWL range gate. Table 8 and Fig. 9 show the quality measures for the altitude analysis.
Statistical measures evaluating the altitude dependency of the wind speed and direction measurement for the two DWLs.
Statistical measures evaluating the altitude dependency of the wind speed and direction measurement for the two DWLs.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0127.1
Due to the limited quantity of available data pairs for WL143, a comprehensive analysis was only feasible for two different altitudes. The upper gate exhibits a statistical scatter σY that is more than twice as high compared to the lower gate, despite having a similar correlation Rυ. Additionally, the systematic difference (BY ≈ 9%) is significantly larger at higher altitudes. The scatter for wind direction σϕ behaves similarly to wind speed, as well as the systematic difference. However, the σϕ values are very close to the 10-min UAS versus tower uncertainty, making it challenging to draw a definitive conclusion.
The larger size of the WL177 dataset allowed a subdivision into more altitude groups: 4 for the wind speed while only 3 for the wind direction due to the additional filtering discarding low wind speed conditions. In the case of wind speed determined by WL177, the statistical scatter σY is comparable for the first three altitudes but shows a clear improvement for the last range: this range is indeed where the WL177 has its focus set to and therefore provides the best SNR. However, this improvement is still much smaller than the scatter between UAS and tower measurements, rendering any further analysis inconclusive. The correlation of wind speed generally increases with altitude, except for a drop to 0.918 in the second altitude range. For the same range gate group, BY reaches a maximum of 9.2% but decreases with increasing altitude. Concerning WL177 wind direction, there is no discernible trend in the quality statistics, although the second altitude group appears to be less accurate.
Simultaneous analysis of altitude effect
Similar to the analysis performed in section 3b(3), again the overall dataset was filtered to compare conditions in which both lidars were operating at the same time. In such a way, a valid comparison of the two systems under the same atmospheric conditions can be exploited and the results are reported in Table 9.
Statistical measures evaluating the altitude dependency for DWLs under the same atmospheric conditions (analogous to the simultaneous comparison).
The two DWLs have similar wind speed measurement quality for the lower altitude group, and there is a tendency toward larger scatter by increasing the scanning altitude. This increase of the statistical scatter seems to be more pronounced for the WL143. For the systematic error, the tendency is opposite for the two systems: WL143 is very accurate in the first altitude, while drifting in the second and vice versa for the WL177.
For the wind direction, measurements of the WL143 have less scatter than the other systems on both altitudes, yet the tendency of increased scatter with altitude is always followed. Regarding the systematic error, the tendency is analogous to the wind speed measurement.
d. Effect of the atmospheric stability
In the last step, our study analyzes whether atmospheric stability has an effect on the lidar wind estimation. To thoroughly examine the impact on the measurements, it was crucial to acquire an ample amount of data under both stable and unstable conditions. However, the WL143 data solely cover IOP-2, restricting the availability of sufficient pairs for stable conditions. Consequently, the investigation of the influence of atmospheric stability was limited to the WL177 system.
1) Richardson number
First, Rib was used as a stability criterion calculated as explained in section 2g. Figure 10 shows the calculated Rib for all the 10-min data pairs. Since the biggest share of flights has been conducted during daytime and in two campaigns in summertime (IOP-1 and IOP-2), more data pairs are for the unstable case than for stable.
Times series of Rib for all the 10-min data pairs available. Values below the critical value of 0.25 denote dynamically unstable conditions, while values higher than 0.25 suggest a stable ABL.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0127.1
The 10-min data were here divided into four groups. The grouping of the positive Rib values represents thermal and dynamic stable conditions (Rib > 0.25) or only thermal stable conditions (0 < Rib < 0.25) (Stull 2016). The subdivision of the negative values was performed by mirroring the boundaries of the positive values simply in order to split the dataset into two groups with significant amount of entries.
2) Lapse rate
In addition to evaluating atmospheric stability from the Richardson number, the same was done with the lapse rate, calculated according to section 2g by using data between 1- and 10-m height above the ground level. By taking the lapse rate as a stability criterion, more data pairs are classified into the stable atmospheric conditions doubling in the end the amount of data in this group. Figure 11 shows the time series of the lapse rate for all the 10-min data pairs. The tendency is similar to Fig. 10; however, more points are above the boundary at zero.
Time series of the lapse rate for all the 10-min data pairs available. Values below 0 denote unstable conditions dominated by convection, while values higher than 0 suggest a stable ABL.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0127.1
3) Stability dependency
By analyzing the statistical measures shown in Fig. 12 and reported in Tables 10 and 11, it is possible to clearly identify an improvement in the DWL wind speed measurement with increasing atmospheric stability. It seems that both values for the bias and the statistical scatter decrease significantly when the atmosphere is stable. This behavior was somehow expected as it is well known that under these specific conditions, the ABL is less active and tends to suppress turbulent motions, resulting in a reduction of the inhomogeneity and nonstationarity of the wind field being measured. However, the scatter is very close to the uncertainty of the UAS itself already in unstable conditions, while for a stable atmosphere, it is way lower. Due to this fact, it is not possible to prove the actual value for the scatter in stable conditions; however, it is plausible that even the uncertainty of the UAS would decrease if calculated only under stable atmospheric conditions. Therefore, the general behavior of these statistical measures is still considered valid.
Visual representation of the statistical measures of the atmospheric stability effect on wind speed and direction measurement for WL177.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0127.1
The wind-direction measurements also show a similar tendency for the statistical scatter. Regardless of the specific value difference between the bulk Richardson number and lapse rate, with stable conditions, the scatter is lower. The tendency is reversed when it comes to the bias where a slightly larger systematic difference is encountered in stable atmospheric conditions.
In conclusion, analyzing the atmospheric stability based on both bulk Richardson number and lapse rate shows that with increasing stability, the WL177 wind vector quality is increasing.
4. Conclusions and outlook
In the framework of the VALUAS project, three field campaigns (IOPs 1–3) during July 2020, June 2021, and November 2022 were carried out at the boundary layer field site Falkenberg, about 60 km southeast of Berlin, in flat but heterogeneous terrain.
The goal of the campaigns and the ongoing data analysis was the validation of Doppler wind lidar (DWL) systems using small uncrewed aircraft systems (UASs) with a special focus on the height range 100–500 m. The latter were operated by the University of Tübingen, while two DWLs running in different modes (WL143 and WL177) were operated by the German Weather Service [Deutscher Wetterdienst (DWD)]. In total, 146 research flights were performed with a typical duration of 1.5 h each. For this study, data from 73 flights were used.
The statistical analysis applied mainly consists of determining systematic differences (bias) B, scatter σ expressed by the root-mean-square difference, correlation coefficients R, and linear regressions in the scatter diagrams.
For both UAS–tower and UAS–DWL validation, the airborne-measured in situ data were subdivided and grouped in a way that data pairs (UAS vs DWL) were as much representative of the same air volume as possible. But of course, instruments like cup anemometers or DWL probe the atmosphere vertically from a fixed location, while aircraft operating on straight and level flight sections (legs) gives horizontal datasets at fixed altitudes. Thus, the two systems do not necessarily probe the same air mass. This factor was considered in the present study by a fetch angle α. For small to medium α (which represents more or less the crosswind share of the airborne measured in situ data), no significant effect on the statistics of both UAS versus tower and UAS versus DWL was found. For α > 40°, the fetch angle effect became noticeable and degraded the statistical agreement.
The first part of the presented analysis validates the UAS data versus tower measurements (cup anemometer and vane) at 98 m AGL. As already shown in the past, see, e.g., Martin et al. (2011), a small research UAS is able to measure—among other quantities—the horizontal wind vector with high accuracy. In the present study, the systematic bias between tower-based cup anemometer and vane data versus the airborne wind vector is about 0.7% (wind speed) and below 4° (direction) for 10-min averaged data. The correlation coefficient between tower and UAS data is better than 0.95 for both wind speed and direction using 10-min averages.
The general comparison of both DWL (WL143 and WL177) versus UAS using the entire database shows that both approaches for measuring the mean horizontal wind velocity υ and direction ϕ yield data in good agreement: correlation coefficients are higher than 0.94 for wind speed and above 0.98 for wind direction. Systematic differences between DWL- and UAS-derived wind direction are both lower than 2.5°. However, systematic overestimation (WL143) of 5.1% and underestimation (WL177) of 5.7% were found regarding the horizontal wind speed.
The grouping of the dataset in altitude ranges (up to 550 m AGL) revealed a dependency of the intercomparison results on height. The WL177 system shows an improvement in the wind speed retrieval quality over altitude, while the wind direction retrieval quality does not show a clear tendency: WL177 is operated with a focus setting at 500 m; this typically implies the best SNR values around this height range. It also seems that the lower range gates contribute more in the underestimation of the wind speed. It should be remembered that the UAS crosses a mixed landscape with some patches of forest in the vicinity of GM Falkenberg. This may have some influence on the wind measurements, especially at the lowest levels. In contrast, the WL143 system shows a decrease in the quality of the wind speed and direction retrieval with increasing altitude, with wind speed being the most affected. Here, on the other hand, it is the higher-range gate that overestimates the wind velocity. Furthermore, up to heights of 500 m, no significant adverse impact of the wider scan cone (resulting from a larger zenith angle) of WL177 is observed in comparison to WL143. This observation holds true despite the fact that the majority of our data was collected within the unstable convective boundary layer.
Atmospheric stability was examined on the basis of the temperature lapse rate γ (i.e., indication of statically stable or unstable conditions) and the bulk Richardson number Rib (i.e., indication of dynamically stable or unstable conditions). Temperature differences at the lower level of the surface layer from 1 to 10 m indicated mostly statically unstable conditions during the first two IOPs in the summers of 2020 and 2021 (γ < 0), while mainly stable conditions were present during the last IOP during autumn of 2022 (γ > 0), as indicated in Figs. 10 and 11. In both cases of wind speed and direction, a general tendency toward a better agreement between the UAS and WL177 is observed through the correlation coefficient, bias, and standard deviation during more stable conditions. Such conditions dampen vertical air exchange and create a more homogeneous, undisturbed wind field, which is closer to the inherent assumption of wind homogeneity that lidar systems often undertake, when they are in operation. This justifies how the overall statistics between the UAS and WL177 improve the more stable conditions, as shown in Tables 10 and 11.
Wind speed and direction differences UAS vs WL177. The four different Rib groups indicate unstable conditions (groups 1 and 2), statically stable but dynamically unstable conditions (group 3), and stable conditions (group 4).
Wind speed and direction differences UAS vs WL177 for thermally stable (2) and unstable cases (1).
Several postprocessing algorithms have been developed and are presently employed for processing data from DWLs operated by the German Weather Service. A forthcoming study aims to investigate the distinctions among these algorithms, utilizing the existing reference UAS data within the altitude range of 100–500 m.
Furthermore, the available database comprises of airborne in situ measurements of turbulence data for the 3D wind vector and air temperature. A subsequent phase of research will be dedicated to the validation of the turbulence variables obtained from DWLs by comparing them against the UAS dataset.
The Doppler lidar toolbox, developed within DWD’s ground-based remote sensing department, is an open source tool to help operators of Doppler lidars, both Streamline and WindCube systems, to process conical lidar measurement in a standardized approach. It takes the native data formats as input, takes care of the prior filtering steps, performs the wind calculation, and outputs the results as netCDF data. The toolbox together with additional information is available at https://github.com/mkay-atm/dl_toolbox.
Acknowledgments.
The authors thank the German Weather Service [Deutscher Wetterdienst (DWD)] for funding and supporting this research, specifically the Federal Ministry for Education and Research for Project Funding (NABF), under grant agreement with a reference: 4819EMF01. The authors declare no conflicts of interest. Jakob Boventer performed all the flights included in the study, did the initial analysis and drafted a first version of the manuscript. Matteo Bramati and Vasileios Savvakis were part of the flight operations, data acquisition, continued and expanded the analysis/discussion, and concluded the writing of the manuscript. Frank Beyrich contributed with guidance on the analysis procedure, discussion and scientific aims of the manuscript. Markus Keyser was involved in writing the description of the lidar systems and assisted in data retrieval from the instruments. Andreas Platis and Jens Bange supervised the flight operations, analysis and discussion, and contributed to the writing of the manuscript.
Data availability statement.
The data used in this work can become available from the corresponding author upon reasonable request.
APPENDIX A
Flight Overview
The complete list of flights performed during the three VALUAS IOPs is reported in Table A1.
List of all MASC-3 measurement times and their contribution to the validation with different timeframes during the IOP-1, IOP-2, and IOP-3 in the years 2020, 2021, and 2022 at the MOL-RAO.
APPENDIX B
UAS Data Processing Flowchart
Figure B1 displays in a schematic way the steps of the UAS data processing.
Schematic representation of the UAS raw data processing procedure. Only the data at the cup anemometer altitude (98 m) are split into three different averaging times. For all the other altitude groups, only 10-min averages are computed and used for WL validation.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0127.1
REFERENCES
Bange, J., and Coauthors, 2013: Measurement of aircraft state and thermodynamic and dynamic variables. Airborne Measurements for Environmental Research: Method and Instruments, A. Kokhanovsky, M. Wendisch, and J.-L. Brenguier, Eds., John Wiley and Sons, 7–75, https://doi.org/10.1002/9783527653218.ch2.
Bange, J., J. Reuder, and A. Platis, 2021: Unmanned aircraft systems. Springer Handbook of Atmospheric Measurements, T. Foken, Ed., Springer, 1331–1349, https://doi.org/10.1007/978-3-030-52171-4_49.
Baumgarten, G., J. Fiedler, J. Hildebrand, and F.-J. Lübken, 2015: Inertia gravity wave in the stratosphere and mesosphere observed by Doppler wind and temperature lidar. Geophys. Res. Lett., 42, 10 929–10 936, https://doi.org/10.1002/2015GL066991.
Chanin, M. L., A. Garnier, A. Hauchecorne, and J. Porteneuve, 1992: Description of a Doppler Rayleigh lidar for measuring winds in the middle atmosphere. Appl. Phys., 55B, 35–40, https://doi.org/10.1007/BF00348610.
Cohn, S. A., and R. K. Goodrich, 2002: Radar wind profiler radial velocity: A comparison with Doppler lidar. J. Appl. Meteor., 41, 1277–1282, https://doi.org/10.1175/1520-0450(2002)041<1277:RWPRVA>2.0.CO;2.
Courtney, M., R. Wagner, and P. Lindelöw, 2008: Testing and comparison of lidars for profile and turbulence measurements in wind energy. IOP Conf. Ser.: Earth Environ. Sci., 1, 012021, https://doi.org/10.1088/1755-1315/1/1/012021.
Davies, B. M., and D. J. Thomson, 1999: Comparisons of some parametrizations of wind direction variability with observations. Atmos. Environ., 33, 4909–4917, https://doi.org/10.1016/S1352-2310(99)00287-3.
Fischler, M. A., and R. C. Bolles, 1981: Random sample consensus: A paradigm for model fitting with applications to image analysis and automated cartography. Commun. ACM, 24, 381–395, https://doi.org/10.1145/358669.358692.
Gasch, P., A. Wieser, J. K. Lundquist, and N. Kalthoff, 2020: An LES-based airborne Doppler lidar simulator and its application to wind profiling in inhomogeneous flow conditions. Atmos. Meas. Tech., 13, 1609–1631, https://doi.org/10.5194/amt-13-1609-2020.
Gottschall, J., M. S. Courtney, R. Wagner, H. E. Jørgensen, and I. Antoniou, 2012: Lidar profilers in the context of wind energy–A verification procedure for traceable measurements. Wind Energy, 15, 147–159, https://doi.org/10.1002/we.518.
Ishii, S., K. Mizutani, T. Aoki, M. Sasano, Y. Murayama, T. Itabe, and K. Asai, 2005: Wind profiling with an eye-safe coherent Doppler lidar system: Comparison with radiosondes and VHF radar. J. Meteor. Soc. Japan, 83, 1041–1056, https://doi.org/10.2151/jmsj.83.1041.
Käsler, Y., S. Rahm, R. Simmet, and M. Kühn, 2010: Wake measurements of a multi-MW wind turbine with coherent long-range pulsed Doppler wind lidar. J. Atmos. Oceanic Technol., 27, 1529–1532, https://doi.org/10.1175/2010JTECHA1483.1.
Klaas-Witt, T., and S. Emeis, 2022: The five main influencing factors for lidar errors in complex terrain. Wind Energy Sci., 7, 413–431, https://doi.org/10.5194/wes-7-413-2022.
Knist, C., V. Lehmann, and M. Kayser, 2018: Das Vorhaben “Pilotstation bodengebundene Fernerkundung”. MOL-RAO Aktuell, 2 pp., https://www.dwd.de/DE/forschung/atmosphaerenbeob/lindenbergersaeule/rao_download/aktuell_2018_02.pdf;jsessionid=895B74A1FF037CAC36DB9BF61BB7A2A7.live11054?__blob=publicationFile&v=1.
Kumer, V.-M., J. Reuder, and B. R. Furevik, 2014: A comparison of lidar and radiosonde wind measurements. Energy Procedia, 53, 214–220, https://doi.org/10.1016/j.egypro.2014.07.230.
Lawrence, T. R., D. J. Wilson, C. E. Craven, I. P. Jones, R. M. Huffaker, and J. A. L. Thomson, 1972: A laser velocimeter for remote wind sensing. Rev. Sci. Instrum., 43, 512–518, https://doi.org/10.1063/1.1685674.
Liu, Z., J. F. Barlow, P.-W. Chan, J. C. H. Fung, Y. Li, C. Ren, H. W. L. Mak, and E. Ng, 2019: A review of progress and applications of pulsed Doppler wind lidars. Remote Sens., 11, 2522, https://doi.org/10.3390/rs11212522.
Lundquist, J. K., M. J. Churchfield, S. Lee, and A. Clifton, 2015: Quantifying error of lidar and sodar Doppler beam swinging measurements of wind turbine wakes using computational fluid dynamics. Atmos. Meas. Tech., 8, 907–920, https://doi.org/10.5194/amt-8-907-2015.
Mariani, Z., R. Crawford, B. Casati, and F. Lemay, 2020: A multi-year evaluation of Doppler lidar wind-profile observations in the Arctic. Remote Sens., 12, 323, https://doi.org/10.3390/rs12020323.
Martin, S., J. Bange, and F. Beyrich, 2011: Meteorological profiling of the lower troposphere using the research UAV “M2AV Carolo”. Atmos. Meas. Tech., 4, 705–716, https://doi.org/10.5194/amt-4-705-2011.
Mengelkamp, H.-T., and Coauthors, 2006: Evaporation over a heterogeneous land surface. Bull. Amer. Meteor. Soc., 87, 775–786, https://doi.org/10.1175/BAMS-87-6-775.
Nechaj, P., L. Gaál, J. Bartok, O. Vorobyeva, M. Gera, M. Kelemen, and V. Polishchuk, 2019: Monitoring of low-level wind shear by ground-based 3D lidar for increased flight safety, protection of human lives and health. Int. J. Environ. Res. Public Health, 16, 4584, https://doi.org/10.3390/ijerph16224584.
Neisser, J., W. Adam, F. Beyrich, U. Leiterer, and H. Steinhagen, 2002: Atmospheric boundary layer monitoring at the meteorological observatory Lindenberg as a part of the “Lindenberg Column”: Facilities and selected results. Meteor. Z., 11, 241–253, https://doi.org/10.1127/0941-2948/2002/0011-0241.
Päschke, E., R. Leinweber, and V. Lehmann, 2015: An assessment of the performance of a 1.5 μm Doppler lidar for operational vertical wind profiling based on a 1-year trial. Atmos. Meas. Tech., 8, 2251–2266, https://doi.org/10.5194/amt-8-2251-2015.
Platis, A., B. Altstädter, B. Wehner, N. Wildmann, A. Lampert, M. Hermann, W. Birmili, and J. Bange, 2016: An observational case study on the influence of atmospheric boundary-layer dynamics on new particle formation. Bound.-Layer Meteor., 158, 67–92, https://doi.org/10.1007/s10546-015-0084-y.
Rahlves, C., F. Beyrich, and S. Raasch, 2022: Scan strategies for wind profiling with Doppler lidar–An Large-Eddy Simulation (LES)-based evaluation. Atmos. Meas. Tech., 15, 2839–2856, https://doi.org/10.5194/amt-15-2839-2022.
Rautenberg, A., and Coauthors, 2019: The multi-purpose airborne sensor carrier MASC-3 for wind and turbulence measurements in the atmospheric boundary layer. Sensors, 19, 2292, https://doi.org/10.3390/s19102292.
Reitebuch, O., and R. M. Hardesty, 2021: Doppler wind lidar. Springer Handbook of Atmospheric Measurements, T. Foken, Ed., Springer, 759–797, https://doi.org/10.1007/978-3-030-52171-4_27.
Robey, R., and J. K. Lundquist, 2022: Behavior and mechanisms of Doppler wind lidar error in varying stability regimes. Atmos. Meas. Tech., 15, 4585–4622, https://doi.org/10.5194/amt-15-4585-2022.
Schön, M., and Coauthors, 2022: Case studies of the wind field around Ny-Ålesund, Svalbard, using unmanned aircraft. Polar Res., 41, 1–15, https://doi.org/10.33265/polar.v41.7884.
Smith, D. A., M. Harris, A. S. Coffey, T. Mikkelsen, H. E. Jørgensen, J. Mann, and R. Danielian, 2006: Wind lidar evaluation at the Danish wind test site in Høvsøre. Wind Energy, 9, 87–93, https://doi.org/10.1002/we.193.
Stull, R., 2016: Practical Meteorology: An Algebra-Based Survey of Atmospheric Science. BC Open Textbook Collection, AVP International, University of British Columbia, 926 pp., https://books.google.de/books?id=xP2sDAEACAAJ.
Teschke, G., and V. Lehmann, 2017: Mean wind vector estimation using the velocity–azimuth display (VAD) method: An explicit algebraic solution. Atmos. Meas. Tech., 10, 3265–3271, https://doi.org/10.5194/amt-10-3265-2017.
van den Kroonenberg, A., T. Martin, M. Buschmann, J. Bange, and P. Vörsmann, 2008: Measuring the wind vector using the autonomous mini aerial vehicle M2AV. J. Atmos. Oceanic Technol., 25, 1969–1982, https://doi.org/10.1175/2008JTECHA1114.1.
Wildmann, N., M. Hofsäß, F. Weimer, A. Joos, and J. Bange, 2014a: MASC–a small Remotely Piloted Aircraft (RPA) for wind energy research. Adv. Sci. Res., 11, 55–61, https://doi.org/10.5194/asr-11-55-2014.
Wildmann, N., S. Ravi, and J. Bange, 2014b: Towards higher accuracy and better frequency response with standard multi-hole probes in turbulence measurement with remotely piloted aircraft (RPA). Atmos. Meas. Tech., 7, 1027–1041, https://doi.org/10.5194/amt-7-1027-2014.
Wildmann, N., S. Bernard, and J. Bange, 2017: Measuring the local wind field at an escarpment using small remotely-piloted aircraft. Renewable Energy, 103, 613–619, https://doi.org/10.1016/j.renene.2016.10.073.
Wildmann, N., N. Bodini, J. K. Lundquist, L. Bariteau, and J. Wagner, 2019: Estimation of turbulence dissipation rate from Doppler wind lidars and in situ instrumentation for the Perdigão 2017 campaign. Atmos. Meas. Tech., 12, 6401–6423, https://doi.org/10.5194/amt-12-6401-2019.
Witschas, B., C. Lemmerz, A. Geiß, O. Lux, U. Marksteiner, S. Rahm, O. Reitebuch, and F. Weiler, 2020: First validation of Aeolus wind observations by airborne Doppler wind lidar measurements. Atmos. Meas. Tech., 13, 2381–2396, https://doi.org/10.5194/amt-13-2381-2020.
zum Berge, K., and Coauthors, 2021: A two-day case study: Comparison of turbulence data from an unmanned aircraft system with a model chain for complex terrain. Bound.-Layer Meteor., 180, 53–78, https://doi.org/10.1007/s10546-021-00608-2.
zum Berge, K., A. Gaiser, H. Knaus, A. Platis, and J. Bange, 2023: Seasonal changes in boundary-layer flow over a forested escarpment measured by an uncrewed aircraft system. Bound.-Layer Meteor., 186, 69–91, https://doi.org/10.1007/s10546-022-00743-4.
