Vertical wind velocity measurements using a five-hole probe with remotely piloted aircraft to study aerosol – cloud interactions

TS1 TS2The importance of vertical wind velocities (in particular positive vertical wind velocities or updrafts) in atmospheric science has motivated the need to deploy multihole probes developed for manned aircraft in small remotely piloted aircraft (RPA). In atmospheric research, lightweight 5 RPAs (< 2.5 kg) are now able to accurately measure atmospheric wind vectors, even in a cloud, which provides essential observing tools for understanding aerosol–cloud interactions. The European project BACCHUS (impact of Biogenic versus Anthropogenic emissions on Clouds and Climate: to10 wards a Holistic UnderStanding) focuses on these specific interactions. In particular, vertical wind velocity at cloud base is a key parameter for studying aerosol–cloud interactions. To measure the three components of wind, a RPA is equipped with a five-hole probe, pressure sensors, and an inertial nav15 igation system (INS). The five-hole probe is calibrated on a multi-axis platform, and the probe–INS system is validated in a wind tunnel. Once mounted on a RPA, power spectral density (PSD) functions and turbulent kinetic energy (TKE) derived from the five-hole probe are compared with sonic 20 anemometers on a meteorological mast. During a BACCHUS field campaign at Mace Head Atmospheric Research Station (Ireland), a fleet of RPAs was deployed to profile the atmosphere and complement ground-based and satellite observations of physical and chemical properties of aerosols, clouds, 25 and meteorological state parameters. The five-hole probe was flown on straight-and-level legs to measure vertical wind velocities within clouds. The vertical velocity measurements from the RPA are validated with vertical velocities derived from a ground-based cloud radar by showing that both mea30 surements yield model-simulated cloud droplet number concentrations within 10 %. The updraft velocity distributions illustrate distinct relationships between vertical cloud fields in different meteorological conditions.


Introduction
Three dimensional wind vectors are an essential parameter for understanding atmospheric processes such as aerosol-cloud interactions and boundary layer turbulence. In tracing the evolution of aircraft-based wind measurements in the atmosphere, three axes of development have been pursued since the 1960s: airborne platforms, inertial navigation systems (INS) and sensors. Airborne platforms have evolved from large aircraft (i.e., Canberra PR3; Axford (1968) or NCAR Queen Air ;Brown 5 et al. (1983)) to ultra-light unmanned aerial systems (i.e., SUMO;Reuder et al. (2008)). INS measure six axes of aircraft motion, and are used to back out wind vectors in the Earth's coordinate frame. A major improvement in INS was the integration of GPS data with fusion sensors (Khelif et al., 1999). The overall accuracy of 3D wind vectors has improved drastically, from 1 m s -1 with wind vanes (Lenschow and Spyers-Duran, 1989) to 0.03 m s -1 with a multi-hole probe (Garman et al., 2006). Since the past decade, GPS, INS and meteorological sensors have become sufficiently miniaturized to be 10 deployed on ultra-light remotely piloted aircraft systems (RPAS) 1 , which has extended observational capabilities previously limited to traditional manned aircraft.
A wide range of RPAS has been used to measure atmospheric winds, from a 30 kg Manta (Thomas et al., 2012) to a 600 g SUMO (Reuder et al., 2008). In particular, a multi-hole probe paired with an inertial measurement unit (IMU; equivalent to 15 INS with integrated GPS data) has been the main mechanism for obtaining 3D winds in fixed-wing RPAS. Ultimately, the combination of multi-hole probe, differential pressure measurements and IMU dictates the precision of atmospheric wind measurements. A lab-built 5-hole probe provides a measurement of vertical velocity within 0.4 m s -1 combined with a GPS-MEMS IMU (van den Kroonenberg et al. (2008); M 2 AV- Spiess et al. (2007); MASC ), and 0.5 m s -1 with a SBG System IMU (ALADINA- Altstädter et al. (2015)). The Aeroprobe, a commercially available 5-hole probe 20 implemented on a Manta platform with a C-Migits-III tactical sensor as IMU, obtained a measurement of vertical velocity to within 0.17 m s -1 on vertical wind (Thomas et al., 2012). A 5-hole Aeroprobe was also used on SUMO RPAS combined with the IMU embedded in the autopilot navigation system to retrieve wind. The uncertainty on 3D wind measurement was provided for the probe reference frame as 0.1 m s -1 (Båserud et al., 2014). A fully-integrated 9-hole probe (with pressure sensors embedded in the probe) has been operated on Manta and ScanEagle RPAS with NovAtel IMU with relatively high 25 precision of 0.021 m s -1 for vertical wind speed and 0.045 m s -1 for horizontal wind (Reineman et al., 2013). Elston et al. (2015) has identified five main points that still need to be addressed for 3D wind measurements using RPAS: (1) true heading remains one of the main sources of inaccuracy in horizontal wind calculation; (2) precise altitude with GPS; (3) miniaturization of IMU for small RPAS, with better accuracy of fusion sensors; (4) improved algorithms for wind field 30 estimation from dynamic soaring; and (5) RPAS regulations and integration in the airspace, which can delay research progress. 1 Commonly called unmanned aerial vehicle (UAV) Until recently, wind measurements from RPAS have been mainly used for the atmospheric boundary layer to study turbulence and atmospheric fluxes. In the BLLAST field campaign, multiple RPAS have been deployed to study the boundary layer during the transition between afternoon and evening periods (Lothon et al., 2014). Results of sensible and latent heat fluxes, and also turbulent kinetic energy (TKE), were calculated from the SUMO RPAS flights,  ;Båserud et al. (2016)). Operation of the M 2 AV and the MASC RPAS during the BLLAST campaign was described in Lampert et al.

5
(2016) with a particular focus on turbulence. TKE decreased along the afternoon-evening transition to reach a minimum near sunset, and turbulence isotropy depended on the presence of a low-level jet. A comparison of near co-located measurements of TKE between different platforms (tethered balloon, RPAS, and manned aircraft) validated the different techniques of obtaining 3D wind vectors (Canut et al., 2016). 10 In addition to the boundary layer studies, aerosol-cloud interactions (ACI) remain one of the main uncertainties in understanding atmospheric processes (Boucher et al., 2013), which is the focus of the collaborative project, BACCHUS (impact of Biogenic versus Anthropogenic emissions on Clouds and Climate : towards a Holistic UnderStanding) (BACCHUS, 2016). The study presented hereafter is part of the BACCHUS project, and presents results from a fleet of RPAS instrumented to study aerosol-cloud interactions. One critical parameter in ACI studies, not previously measured by RPAS, is 15 the vertical velocity w at cloud base, which has been identified as essential to quantifying the aerosol effect on cloud properties (Hudson and Svensson (1995); Snider and Brenguier (2000); Schmidt et al. (2015)). Peng et al. (2005) showed the importance of vertical velocity for convective clouds in a cloud closure study, and highlighted the need of more cloud microphysical data to further test the sensitivity of cloud droplet number concentration to variations in vertical velocity. In Conant et al. (2004) and Sanchez et al. (2017), updraft has also been described as a critical parameter, along with cloud 20 condensation nuclei (CCN) spectra, to derive cloud droplet number concentration (CDNC) in ACI studies.
The motivation of this present work is to assess the ability of RPAS to measure vertical wind velocity near cloud base to study aerosol-cloud interactions. The first sections of the manuscript describe the RPAS platform and the methods used to calculate 3D winds. Then, the details of calibration of the 5-hole probe in a wind tunnel are discussed, complemented by an 25 uncertainty analysis on vertical wind velocity, w (the main parameter needed for ACI studies). A comparison with a sonic anemometer on a meteorological mast provides a validation of RPAS measurements in relatively calm wind conditions. Lastly, several case studies focus on flights under a range of turbulent conditions, during a BACCHUS field campaign in Ireland, and vertical wind velocities from the RPAS are compared to those of a cloud radar. The RPAS used here to measure 3D winds and study aerosol-cloud interactions are based on the commercially available Skywalker X6 model. The wingspan is 1.5 m long, and take-off weight varies between 1.5 kg and 2.3 kg depending on the mission specific payload. The navigation system is the open source autopilot Paparazzi from Ecole Nationale de l'Aviation 5 Civile (Brisset et al., 2006). One of RPAS (wind-RPAS) is specially equipped to measure three dimensional wind vectors, whose validation and study of different cloud cases is the purpose of this work. Its take-off weight is 1.5 kg for a payload of 500 g with cruise airspeed approximately 16 m s -1 .

Instrumentation
The payload of the wind-RPAS to measure 3D wind vector is composed of temperature (IST, Model P1K0.161.6W.Y.010), 10 pressure (All Sensors, Model 15PSI-A-HGRADE-SMINI) and relative humidity sensors (IST, P14 Rapid-W). Two Licor LI-200R pyranometers are installed on the fuselage; one facing up to measure downwelling solar irradiance, and the other facing down to measure upwelling solar irradiance, and are used to detect presence of cloud. Wind vectors are obtained from a 5-hole probe linked to its pressure sensors (All Sensors) by tubing, and an Inertial Measurement Unit, IMU (Lord Sensing Microstrain 3DM-GX4-45). The data from both the IMU and the pressure sensors are recorded by the same acquisition system to ensure 15 precise synchronization. The acquisition frequency is 30 Hz, and data are averaged to 10 Hz for analysis. The 5-hole probe is constructed by the Aeroprobe Corporation, and consists of stainless tube with a semi-spherical tip (Fig.1). The associated electronics have been designed at the Centre National de Recherches Météorologiques (CNRM) laboratory, and consist of three differential pressure sensors (All-Sensors 5inch-D1-MV) and one absolute pressure sensor (All Sensors MLV-015A). Figure   1a illustrates the probe schematic: hole 1 measures the total pressure; the differential pressure between holes 2 and 3 provides 20 β, the angle of sideslip; the differential pressure between 4 and 5 gives α, the angle of attack; and hole 6, a ring around the probe, corresponds to the static pressure port. The difference between total pressure (hole 1) and static pressure (hole 6) gives the dynamic pressure, and determines the airspeed, V a . To obtain angles in degree and airspeed in m s -1 , the 5-hole probe system must be calibrated in the probe's coordinate system and converted to the Earth's coordinate system. The IMU sends information to the acquisition system regarding attitude angles, roll φ, yaw ψ and pitch θ, GPS time and GPS position and 25 altitude, and ground speed of the RPAS in Earth's coordinate system. Schematics of coordinate systems and angles are shown in Fig.1b and also described in Lenschow and Spyers-Duran (1989), Boiffier (1998)

Methods
3D wind vector in the Earth's coordinate system is obtained by subtracting the measured motion of the plane (given by the IMU), from the motion of the air (given by the 5-hole probe). The measurement of 3D winds involves the fusion of three in the probe coordinate system; while the attitude angles θ, ψ and φ are given in the RPAS coordinate system. Lenschow equations (Lenschow and Spyers-Duran, 1989) are then used to calculate the wind vector in the Earth's coordinate system. The angular acceleration of the RPAS is negligible, because the distance between the 5-hole probe and the IMU is on the order of centimeters. In addition, we only consider data from straight and level flight for the study here, which simplifies the Lenschow equations to: The 3D wind vectors u, v, w are given in the Earth's coordinate system in Eq.(1), and are the three components of the wind on x, y and z-axis, respectively (Fig.1b). The positive x-axis represents East, the positive y-axis represents North, and the 10 positive z-axis represents upward direction. V e , V n , V p are provided by IMU, and are the RPAS ground velocities along the x, y and z-axis. ψ and θ are the yaw and the pitch angle, respectively, determined by the IMU. α is the angle of attack, β is the angle of sideslip, and V a represents the airspeed, provided by the 5-hole probe, and initially measured in the 5-hole probe coordinate system. Equation (1) takes into account rotations between the different coordinate systems to eventually provide 3D wind vectors in the Earth's coordinate system. with a cross section of 70 cm, and a wind range between 0.15 to 50 m s -1 . The uncertainty associated to the wind velocity in the wind tunnel is less than 2 %. The calibration of the 5-hole probe is a two-step process -first establishing the relationship between the absolute pressure sensor, raw voltage to mbar with a barometer, then associating the differential pressure in mbar 20 to angles (α and β) or velocity (V a ) in degree or m s -1 , respectively. The probe, its electronics, and the IMU are installed on a two-axis platform with motion in vertical and horizontal planes. The multi-axis platform rotates in the pitch axis (motion in the vertical plane) and yaw axis (motion in the horizontal plane), controlled with a LabView program (Fig.2). The amplitude of pitch and yaw angles varies from -15 deg to 15 deg to simulate flight conditions.

Static calibration
The angle of attack α and the angle of sideslip β are obtained from linear relationship between IMU angles and ratios of differential pressure sensors. The ratios are defined by the following relationships: ∆(P 2 −P 3 ) is the differential pressure between holes 2 and 3, related to the calculation of the angle of sideslip β; ∆(P 4 −P 5 ) is the differential pressure between holes 4 and 5, related to the calculation of the angle of attack α; and ∆(P 1 − P 6 ) is the differential pressure between holes 1 and 6, related to the airspeed. The linear relationship between C α and C β (5-hole probe) and the yaw angle ψ and the pitch angle θ (IMU) determines the calibration coefficients (Fig.3).

5
Cα 1 = m 1 × (α + α 0 ) + n 1 and Cα 2 = m 2 × (−α + α 0 ) + n 2 (3a) The linear calibration coefficients are denoted by m, n, j, k while α 0 and β 0 are offsets in α and β associated to the alignment of the pressure ports on the probe. In the calibrations performed here (Fig.3), α 0 and β 0 are found to be -0.76 deg 10 and -0.84 deg, respectively. Subscript 1 corresponds to measurements with the probe in its standard orientation, while subscript 2 corresponds to measurements with the probe in its inverted orientation (i.e., rotated 180 deg). To calculate the calibration coefficients m and n for the angle of attack, the yaw angle was held constant to zero while the pitch angle varied, and vice versa to obtain the calibration coefficients k, and j for the angle of sideslip. In the wind tunnel, the α (5-hole probe) and the θ (IMU) angles are the same, as for β (5-hole probe) and ψ (IMU) angles. To account for offsets in the alignment, 15 experiments are performed with the probe in the standard orientation shown in Fig.1, with roll angle equal to 0 deg (subscript 1), and the probe in inverted orientation for the roll angle equal to 180 deg (subscript 2). Likewise, the same procedure is followed with 90 deg and -90 deg to determine β 0 . IMU angles and ratios of differential pressure sensors are recorded for platform positions between -15 deg and 15 deg for three air speeds in the wind tunnel. Figure 3 shows that calibration coefficients do not change for the range of airspeeds in these experiments (between 15 and 25 m s -1 ).

20
In an experiment to verify linearity between the calibration coefficients (C α , C β ) and the range of angles (α, β) supported for flights, the pitch and yaw angles of the multi-axis platform ( Fig.4) are varied concurrently. Figure 4a shows the applied values of IMU pitch and yaw angles to the platform, and Fig.4b illustrates the corresponding values of C α , C β from the 5-hole probe. The experiment is conducted at a constant wind speed of 15 m s -1 . The slight angular bias in Fig.4 is due to a 3 25 degree roll angle in the mounting of the two-axis platform in the wind tunnel. In Fig.4b, non-linearity is observed when C α or C β exceed ±1, which corresponds to α and β angles between ±10 deg. Thus Where γ = 1.4 is the specific heat ratio, R = 287.07 J kg -1 K -1 , T is the temperature in Kelvin, P t is the total pressure in mbar, and P s is the static pressure in mbar. The absolute pressure at the tip of the 5-hole probe (hole 1; Fig.1a) is P t = ∆(P 1 − P 6 ) + P 6 . The static pressure is measured at hole 6 with P s = P 6 . The probe airspeed was calibrated using the wind tunnel between 12 and 34 m s -1 velocities. Equations (4) account for temperature T and absolute total pressure P t , and are within 7 % of nominal value. The airspeed calculated with the Bernoulli's law yielded similar results within 10 %.

Dynamic calibration
After performing the static calibration (Section 3.1; Fig.3 and Fig.4), the validation of the 5-hole probe is conducted in the wind tunnel, imposing a triangular motion of the pitch angle with different amplitudes and frequencies (Fig.5, inset). The wind tunnel provides a laminar flow, thus updrafts and downdrafts are, by definition, negligible even when the multi-axis platform is in motion. The results in Fig.5 show that w averages 0 m s -1 (uncertainty analysis in the next section), which indicates that the 15 platform motion has successfully been removed. However, the standard deviation of w (1-σ) increases notably with the rate of change of the pitch angle of the platform (Fig.6). Under the flight conditions reported in this work (Tables 2 and 3), the pitch angle rarely exceed ±10 deg s -1 (Section 3.1). As the focus of the scientific research related to these wind measurements centers around updraft observations for aerosol-cloud interactions studies, Section 3.3 describes the uncertainty analysis related to the calculation of vertical wind w.

Uncertainty analysis
The uncertainty analysis on the vertical wind vector w, Eq.(5), shows that the standard deviation, σ w , is less than 0.1 m s -1 based on wind tunnel experiments (Fig.5), with a lower limit of 0.05 m s -1 for a fixed probe (no motion). As shown in Fig.6, the magnitude of σ w depends on the rate of angular change of the pitch angle, which suggests that σ w is related to a convolution induced by the signal processing in the IMU that measure accelerations and angles. Figure 6 shows the σ w 25 increases with higher frequency motion. In this section, we explore the parameters related to the uncertainty analysis for the vertical wind vector w. Equations (5) and (6) where the relationship between measurement of differential pressure on the 5-hole probe and the corresponding α angle is given by: The uncertainties associated with each parameter from Eq.5 and Eq.6 are summarized in Table 1 and show that the lower limit for measuring updraft with this system is 0.11 m s -1 corresponding to the observed measurements in Fig.6. to calculate turbulent kinetic energy (TKE) used to study boundary layer dynamics and to compare with previous studies.
15 Table 2 (Table   2). While all flights were conducted in low wind conditions, the turbulent conditions differed from one flight to another.

Validation of 3D wind measurements
Once the three components of the wind velocity from the RPAS are obtained with the Lenschow equations, Eq. (1) anemometer were averaged (Fig.7). The PSDs transform the wind components into a frequency domain, and reveal the contribution of the RPAS in the wind velocity components. Note that frequencies lower than 10 -2 Hz are sparse, therefore, the averaged values appear less smooth than higher frequencies. For flights 1, 2, 3 and 4, the RPAS motions are still visible at 0.1 Hz in the horizontal u-component due to inaccuracy of the heading measurement in the IMU. Flight 5 was conducted after improved calibration of the heading measurement by the IMU, and a notable improvement in the PSDs ( Fig.7; green line 5 particularly on u-component) clearly demonstrates the importance of precise heading measurements. The PSDs of the wind velocities from both RPAS and anemometer follow the -5/3 slope as expected from the Kolmogorov law. We note, however, that the PSD from the RPAS is higher than the wind from anemometer, owing to the motion of the RPAS platform. The same trend has also been observed with a Manta UAV by Reineman et al. (2013) and a SUMO platform at Lannemezan in 2011 (Reuder et al., 2008).

Turbulent kinetic energy
In the atmospheric boundary layer, the turbulent kinetic energy (TKE) quantifies the intensity of turbulence, which controls mixing of the atmosphere (Wyngaard and Coté (1971); Lenschow (1974)). TKE is defined as:  Fig.10 shows that σ v,mast 2 is up to a factor of two higher than σ u,mast 2 ; therefore isotropic conditions are not satisfied at 30 m.agl. While at 60 m.agl, σ u,mast 2 and σ v,mast 2 are within 15 %, implying isotropy for altitudes in the boundary layer above 60 m.agl. Differences observed in variances between the horizontal wind components at 30 m.agl are related to surface topography (e.g., nearby trees, fields. . . ).

30
While isotropic conditions are satisfied for measurements in the boundary layer above 60 m.agl, based on mast observations, these expected conditions are not observed between the transversal direction (normal to the axis of the RPAS fuselage) and longitudinal direction (parallel to the axis of the RPAS fuselage) based on the RPAS measurements. Therefore, to calculate TKE RPAS , we assume isotropy, where σ u,RP AS 2 equals σ v,RP AS 2 . Figure 9 illustrates the impact of the isotropy In Fig.9, the TKE calculated for flight 4 is significantly different from the other flights as the RPAS was close to stall speed (as mentioned in Section 4.1, flight 4 also yielded the lowest intersection number). While beyond the scope of this paper, these results suggest that improving the measurement of horizontal winds and reducing biases in the horizontal components of 10 the variances may be achieved by 1) improvement of the IMU heading measurement (also noted in Elston et al. (2015)), and 2) verified with a flight plan in a cross pattern (i.e., orthogonal legs).

Comparison of vertical wind velocities from RPAS and cloud radar
A BACCHUS field campaign took place at the Mace Head Atmospheric Research Station on the west coast of Ireland in 15 August 2015. The purpose was to study aerosol-cloud interactions linking ground-based and satellite observations using RPAS (Sanchez et al., 2017). Among the four instrumented RPAS which flew at Mace Head, the wind-RPAS was equipped with a 5-hole probe and an IMU to obtain 3D wind vectors, as well as upward and downward facing pyranometers to measure downwelling and upwelling broadband solar irradiance (400 to 1100 nm wavelengths). During the campaign, we concentrated on measurements of vertical wind velocity near cloud base to study aerosol-cloud interactions. After identifying the cloud 20 base from the ceilometer or a vertical profile of an earlier flight, the wind-RPAS was sent to an altitude close to cloud base, flying 6 km-long straight-and-level legs. Horizontal wind speeds varied from 6 to 12 m s -1 from the West during the case studies presented here. The presence of clouds during the flight was determined using the ratio of the upwelling to downwelling solar fluxes; when the RPAS was underneath or within a cloud, the ratio approaches unity. During this field campaign, the wind-RPAS flew in 10 of the 45 scientific flights for a total of 15 hours. Here, we focus on three flights with the cloud radar is selected for the comparison; either cloud top or RPAS flight altitude. The vertical resolution of the cloud radar is 29 m. While the flight altitude for the wind-RPAS was estimated to be near the cloud base, uncertainties in retrieving cloud base height or an evolution in cloud base height related to diurnal cycles of the boundary layer inevitably lead to the possibility of the RPAS flying in the clouds rather than just below cloud base. Consequently, data from two wind-RPAS flights during the field campaign are not presented due to the accumulation of water in the probe. We are currently improving the 5 instrumentation to address the issue. Note that direct comparison of instantaneous data between the RPAS and cloud radar was not possible, as the RPAS did not fly directly over the cloud radar, and did not observe the same air mass. Moreover, the cloud radar reports vertical velocities every 10 s (and only when a cloud is present); therefore relatively long averaging preriods are needed to compare with the RPAS observations. Hence, we present selected time series of the cloud radar measurements that represent the state of the atmosphere during the flight; for cases with sufficient cloud cover, we present 10 different averaging periods of the cloud radar (a short period that coincides with the RPAS flight and a long periods for better counting statistics). Normalized distributions are plotted on the same interval divided into 30 bins of vertical wind velocities.
In the present study, comparisons of the vertical wind velocity of the cloud radar serve to validate the RPAS results, as well as to provide insight on different atmospheric states related to the measurement techniques.  2015)). These negative biases related to the falling drops are largely removed by obtaining vertical velocity at the top of the cloud . Similar results are obtained for our case study, as the cloud radar is strongly influenced by falling droplets, yet only slightly negatively biased at the cloud top. The intersection method, described in Section 4.1, is used to compared the three radar vertical wind distributions with the distribution of RPAS measurements (Table 3). The intersection number of 0.53 between radar flight 30 altitude and RPAS vertical wind distributions confirms the low match as a result of the negative bias from the precipitating droplets. However, a much better agreement is found between the cloud radar vertical velocity at the top of the cloud (1360 m.asl) and the RPAS measurements (intersection number = 0.74). A comparison of results from the RPAS and cloud radar emphasizes the differences in vertical winds depending on the regions within the cloud field. During Flight 38, the RPAS flew within a cloud above the ocean and in clear sky above land for three legs, after which the local meteorology changed into a formation of developing clouds above land (where a cloudless sky had previously been observed; Fig.13). The vertical wind velocity for the Flight 38 is presented using a combination of 5 information shown in a series of figures: downwelling and upwelling pyranometer observations, and three periods corresponding to distinct meteorological conditions (Fig.13); the time series cloud radar data (Fig.14); and the vertical wind distributions from the RPAS flight and the cloud radar (Fig.15). These meteorological periods are defined in Fig.13 as "cloud" (both pyranometers approach similar values), "no cloud" (downwelling pyranometer is significantly higher than upwelling pyranometer), and a third period associated to a developing field of broken clouds (spatially variable downwelling 10 pyranometer). Combining information from Figs.13, 14, and 15, we deduce a cloudless sky (cyan) was observed by the RPAS above land for the first three legs (Fig.13). The corresponding cloud radar time series also showed a cloudless sky above land for the beginning of the flight (Fig.14). In the meantime, the RPAS flew within a cloud above the ocean (green), which was not observed by the cloud radar. Figure 15 shows that the standard deviation of vertical velocity within the cloud is larger than for clear sky conditions (σ cloud = 0.29 m s -1 , σ no cloud = 0.17 m s -1 ) , which highlights the presence of stronger vertical winds 15 in the presence of clouds. During the last two legs of Flight 38, the wind-RPAS flew through a developing field of broken clouds above land (magenta), which also appeared in the cloud radar time series and in the satellite image ( Fig.4 in Sanchez et al. (2017)). The standard deviation of the "broken cloud" RPAS period of vertical wind is larger than the other periods (σ broken cloud = 0.48 m s -1 ). Similar vertical wind distributions are found for cloud radar and the RPAS during the "broken clouds" period ( Fig.15). While not shown here, the vertical wind distributions observed by the cloud radar are similar at radar 20 cloud base (380 m.asl) and at the flight altitude (660 m.asl), as well as at different observing periods (1.5 and 4 hours). In Table 3, intersection numbers illustrate the relatively close matches (ca. 80 %) in comparing the "broken cloud" RPAS period and the cloud radar for 4 hours (radar flight altitude) and for 1.5 hours (radar flight time). For comparison, the values of intersection numbers between "broken cloud" and "cloud" periods is 0.73, while between "broken cloud" and "no cloud" periods is 0.56 (based on RPAS measurements). The similar results for the observations of a field of broken clouds 25 independently reinforces RPAS and cloud radar observational methods, and the changes meteorological conditions highlight the ability to identify distinct states of the atmosphere with the RPAS. Relating these differences in updraft velocity to the meteorological conditions of the boundary layer will be explored in future studies. distribution, therefore only the cloud radar data for 4 hours (red segment) are presented in Fig.18. To compare cloud radar and RPAS data, vertical winds from the RPAS are divided into "cloud" and "no-cloud" periods based on pyranometer observations. The respective standard deviations for the periods are σ cloud = 0.37 m s -1 and σ no cloud = 0.36 m s -1 , which are not statistically different. However, the variability between legs is significantly greater in the "no cloud" period (as represented by the envelope in blue dashed lines in Fig.18) compared to the "cloud" period (envelope in green dashed lines).

5
In Fig.18, the RPAS and cloud radar measurements show similar results during the "cloud" period, with an intersection number equal to 0.78. Kunz and de Leeuw (2000)

Conclusions
The validation of 3D wind measurements measured by a 5-hole probe on a lightweight remotely piloted aircraft system (RPAS) 15 has been detailed in this study. The 5-hole probe has been calibrated in wind tunnel on a dynamic platform to obtain the angle of attack, angle of sideslip and airspeed of the RPAS. Motions induced by the two-axis platform in the wind-tunnel were effectively removed, thereby validating sensor performance. With an inertial measurement unit (IMU) providing ground speeds, Euler angles and GPS coordinates, 3D wind vectors have been calculated with the simplified wind equations from Lenschow and Spyers-Duran (1989). The uncertainty associated with the vertical wind measurement has been determined to 20 be 0.11 m s -1 . The 3D wind vectors from the RPAS showed good agreement with results from a sonic anemometer on a 60 m.agl meteorological tower at P2OA, Lannemezan, France. Vertical velocity distributions were compared from both platforms, and showed intersection values higher than 70 % in calm wind conditions. Comparisons have also been made on the power spectral density (PSD) functions between the sonic anemometer and RPAS measurements, which in both cases follow the Kolmogorov law for established turbulent regime. In order to calculate the turbulent kinetic energy (TKE) parameter, the isotropy assumption 25 (σ u 2 = σ v 2 ) has been applied on the horizontal wind from RPAS in order to correct biases in measurements resulting from heading inaccuracy. In the future, heading measurements will be improved with an IMU that includes differential GPS antennas  Table 1. Uncertainty (1-σ) associated with parameters from 5-hole probe (5HP) and inertial measurement unit (IMU) for the calculation of error associated with the vertical wind velocity w.

Variable
Symbol precision/value differential pressure between holes 1 and 6 (5HP) σ ∆(P 1 −P 6 ) 0.012 mbar differential pressure between holes 4 and 5 (5HP) σ ∆(P 4 −P 5 ) 0.012 mbar ratio of differential pressures (   Δ(P 1 -P 6 ) Air speed Va Δ(P 4 -P 5 ) Angle of attack α Δ(P 1 -P 6 ) Angle of sideslip β P 6 Absolute static pressure P 1 =P 6 +Δ(P 1 -P 6 ) Total pressure        Upwelling pyranometer Downwelling pyranometer Figure 13. Coastal map and flight tracks for the case study of a convective cloud with changing meteorology (Flight 38). Downwelling and upwelling pyranometers data are color-coded based on the three flight periods ("cloud", "no cloud" and "broken clouds"). The developing field of broken clouds (magenta contour clouds) appeared during the last two legs. The cloud radar (yellow square) operated at the Mace Head research station.