2.1 University of Colorado Boulder RAAVEN

The University of Colorado Boulder RAAVEN (Fig. 1) is a fixed-wing sUAS that has been developed for the collection of detailed information on the structure of the atmosphere and has been operated by the University of Colorado team since 2019. The RAAVEN airframe is based on the commercially available DRAK UAS manufactured by RiteWing RC and has a wingspan of 2.3 m. The airframe has been updated to meet the needs of atmospheric science missions spanning a variety of environments. The RAAVEN uses the Pixhawk2 flight controller and is powered by an 8S 21 000 mAh lithium ion (Li-Ion) battery pack to offer flight times around 2.5 h with minimal payload. The airframe was modified to include a tail boom in order to assist with improvement of longitudinal stability and overall performance. The aircraft has a maximum airspeed of approximately 36 m s⁻¹, though during TRACER flights were generally conducted in the 16–19 m s⁻¹ range.

Figure 1The RAAVEN UAS, as instrumented in the field for TRACER.

Download

For TRACER, the RAAVEN carried sensors from the miniFlux payload co-developed by the National Oceanic and Atmospheric Administration (NOAA), the Cooperative Institute for Research in Environmental Sciences (CIRES), and Integrated Remote and In Situ Sensing (IRISS) at the University of Colorado. This features a primary suite of instruments (see Fig. 1), specifically including a pair of RSS421 PTH (pressure, temperature, humidity) sensors from Vaisala, Inc., a multihole pressure probe (MHP) from Black Swift Technologies, LLC (BST), a pair of Melexis MLX90614 IR thermometers, a custom fine-wire array developed and manufactured at the University of Colorado Boulder, and a VectorNav VN-300 inertial navigation system (INS).

In addition to the sensors described above, the RAAVEN carried a Printed Optical Particle Spectrometer (POPS, Gao et al., 2016), developed by the NOAA Chemical Sciences Laboratory and currently sold commercially by Handix Scientific. POPS provides information on aerosol size distribution, giving size-resolved concentrations across 24 size bins as well as information on a variety of instrument monitoring systems. POPS data are available for all but four of the completed flights, during which sensor overheating resulted in no usable data. This overheating issue was resolved by adding additional ventilation to the payload bay carrying the POPS sensor once it was discovered. POPS data quality has been evaluated in previous studies (e.g., Mei et al., 2020). The inlet for the POPS instrument was located on the RAAVEN fuselage and consisted of a 2 mm ID (3 mm OD) brass tube mounted in an isoaxial configuration. POPS was operated at a sample flow rate of 3 cm³ s⁻¹, resulting in sub-isokinetic sampling at RAAVEN airspeeds and leading to size-dependent oversampling of particles. The sample line between the inlet and the instrument was constructed of brass and conductive silicone tubing. Overall particle sampling efficiency accounting for aspiration and transmission (Baron and Willeke, 2001) ranged from 1.1 to 2.1 over the POPS measurement size range. Data quality can additionally be impacted by a variety of different things, including sensor temperature and flow rate. These issues are diagnosed in postprocessing, and the POPS data quality flag (see below) helps provide users with additional information on sensor data quality.

With this combination of sensors, the RAAVEN was configured to observe the atmospheric and surface properties necessary for evaluating kinematic and thermodynamic states, turbulent fluxes of heat and momentum, and the aerosol size distribution for particles between 150 and 2500 nm. The addition of this sensor suite reduced the aircraft endurance to approximately 90 min, as the POPS installation took up some of the space normally allocated for batteries. All the sensors along with aircraft autopilot data were logged using a custom-designed FlexLogger data logging system. Detailed information on the performance of the different sensors and data acquisition rates can be found in de Boer et al. (2022) and Cleary et al. (2022) and is therefore not repeated here.

Figure 2The CopterSonde UAS with a radio controller.

Download

2.2 University of Oklahoma CopterSonde

The CopterSonde is a rotary-wing quadcopter used to collect frequent vertical profiles of the ABL (Fig. 2). It was developed at the University of Oklahoma and is maintained by the Cooperative Institute for Severe and High-Impact Weather Research and Operations (CIWRO). The platform is 0.5 m in diameter and weighs 2.3 kg, making it easily transportable. The CopterSonde uses a combination of direct sensors, autopilot software, and algorithms to gather a profile of atmospheric data. Pressure, temperature, and humidity are observed using an MS5611 barometric pressure sensor, iMet-XF bead thermistor, and HYT-271 humidity capacitor, respectively. The pressure sensor is integrated into the Pixhawk CubeOrange autopilot board to improve the altimeter estimation. To remove temperature fluctuations from the pressure observations, the Pixhawk is heated to a constant temperature within the first 2 min of start-up. The thermistors and humidity capacitors are housed in the intake scoop of the CopterSonde, where they are sheltered from insolation and heat from the motors or Pixhawk Cube. The positioning of temperature sensors was selected based on findings in Greene et al. (2018). Within the intake scoop, there is a small fan to aspirate the sensors, although it does not turn on until the CopterSonde reaches an elevation of 3 m above ground level (a.g.l.), to avoid ingesting dust into the scoop. In addition to the fan aspiration, the autopilot implements the wind vane mode explained in Segales et al. (2020) to direct the scoop into the prevailing flow. As such, the air is not altered by the UAS before it passes over the sensors. Additionally, the wind vane mode improves wind speed and direction estimation by increasing axis symmetry and reducing vibrations. Wind speed and direction are determined by a linear algorithm estimator using aircraft attitude described more in Sect. 4.2. Sensor accuracy response times and further specifics on the system specifications can also be found in Segales (2022).

Figure 3TRACER-UAS flight locations over a map of the Houston–Galveston area with zoomed-in inlays of the flight paths by the RAAVEN (blue line) and profiling site (red dot). The right and left maps are satellite imagery courtesy of © Google Maps 2022. The center map uses data from © OpenStreetMap contributors 2023. Distributed under the Open Data Commons Open Database License (ODbL) v1.0.

4.1 University of Colorado Boulder RAAVEN

Data collected by the RAAVEN's sensors during TRACER-UAS were logged at a variety of different logging rates. As with previous deployments, the fine-wire system was logged at 250 Hz, the fastest rate of all of the sensors. The BST MHP was logged at 100 Hz, the VectorNav VN-300 at 50 Hz, the Melexis IR sensors and variables related to fine-wire status at 20 Hz, data collected from the Pixhawk autopilot and Vaisala RSS421 sensors at 5 Hz, and data from the POPS aerosol spectrometer, a new addition for this campaign, at 1 Hz. All logging events carried out by the FlexLogger include a sample time from the logger CPU clock, allowing for post-collection time alignment between the different sensors. A detailed description of the time alignment process is included in Cleary et al. (2022).

For the B1-level data files, the resampled (in time) data include several derived and measured quantities, which are provided at 10 Hz. These data include information on aircraft position, including information on latitude, longitude, and altitude, as measured by the VN-300. Aircraft altitude is corrected using a combination of various inputs from onboard GPS and pressure altimeters, as neither of these altitude estimates can be used reliably as a definite flight altitude. Information on derivation of the aircraft altitude is also provided in Cleary et al. (2022). As with previous campaigns, a flight_flag binary variable is developed by combining information on aircraft airspeed and altitude, as provided by the autopilot system. Times when the aircraft airspeed exceeds 10 m s⁻¹ and the aircraft altitude is greater than 5 m a.g.l. are flagged as periods when the RAAVEN is flying (flight_flag = 1). The time point 4 s (200 samples) before the first point where flight_flag is set to 1 is recorded as the takeoff point, while the time point 4 s (200 samples) after the last flying point in the record is designated the index where the aircraft has landed.

Figure 5Typical flight altitude patterns from the RAAVEN (blue) and CopterSonde (red) during a single RAAVEN flight. Dashed boxes highlight examples of vertical profiles pulled from each platform for the data comparison.

Download

Deriving wind information from fixed-wing research aircraft systems is a complex undertaking (see van den Kroonenberg et al., 2008). Doing so requires a combination of information from different sensors, including measured airspeed, airflow angle over the aircraft, and aircraft motion relative to the Earth system. For the RAAVEN platform, any biases in true airspeed (TAS) can impart significant errors in the calculation of wind velocity, while time lags between the reported GPS velocities and in situ measured aircraft attitude and any angular offsets between the INS and MHP tend to have smaller impacts. In this study, these potential sources of error are corrected for by implementing an optimization technique. In this technique, small adjustments are made to individual parameters, including airspeed, angle of attack, sideslip angle, and temporal logging offset, to generate a wind solution. Then these individual wind solutions are evaluated, and the one with the smallest overall sinusoidal variability over individual orbits or racetracks is selected as the correct combination for deriving wind parameters (see Cleary et al., 2022, for full details).

To improve the usability of the parameters measured during the TRACER campaign, the data files developed as part of the RAAVEN dataset have been assigned data quality flags. These flags are determined through a variety of means, as described here. The flag associated with the RSS421-derived temperature is set to 0 for time periods that are deemed to consist of good data and is set to 1 for times when there are potential data quality issues, as identified by (a) the absolute value of the difference between the temperature from either individual sensor being greater than 0.5 °C, (b) the absolute value of the difference between the RSS421 temperature and the temperature from the EE-03 sensor on the MHP exceeding 5 °C, (c) the internal error flag of either RSS421 sensor being active, or (d) the aircraft not being in the flying state identified using the flight_flag parameter. For the RH (relative humidity) measurement from the RSS421, similar criteria are implemented, except that limits are set to be 5 % between RSS421 sensors and 15 % between the output RH value and the MHP-provided RH value. This second value is as large as it is because the RH values from the MHP-mounted sensor are impacted by exposure of that sensor to sunlight and the associated impact on sensor temperature. Because these temperature swings are not corrected for, this MHP-mounted sensor can produce large fluctuations in the RH values. As a result, this MHP-based RH measurement is only meant to provide a reality check to ensure that the RSS421 sensors are reporting accurate values. The most important comparison is between the two RSS421 sensors, which should agree much more closely, as they are the same sensor type and are mounted in close proximity to one another.

In addition to the RSS421 flags, there is also a data quality flag implemented for the cold-wire temperature sensor. This data quality flag is activated when the difference between the cold-wire-derived temperature value and either RSS421 temperature exceeds 0.6 °C, when the absolute value of the difference between the cold-wire-derived temperature and that from the MHP-mounted sensor exceeds 2 °C, when cold-wire voltages are observed to fall outside of the 0–4 V analog range, or when flight_flag is 0. There is also a pressure quality control flag for the pressure measurements from the VN-300. This flag is activated if the absolute value of the difference between the VN-300 static pressure and that measured by either RSS421 sensor exceeds 4 hPa. The RSS421 pressure measurements are not used as the primary pressure measurement because comparisons with radiosonde and tower data indicate that they are likely biased low due to the airflow passing over their location on the aircraft.

The RAAVEN dataset also includes a three-stage data quality flag for wind estimates from this platform. This flag is set to 0 for times where wind data are deemed to be good, 1 for time periods where data are potentially suspect, and 2 where data are known to be of poor quality. Data are labeled “bad” if any of the following are met.

• The measured angle of attack or sideslip exceeds 20°. Times where the angle of attack or the sideslip are between 10 and 20° are flagged as “suspect”. This is because the multihole probe is calibrated to an ±15° angle of attack or angle of sideslip.

• The measured TAS is less than 10 m s⁻¹.

• There is noted blockage of MHP ports, as indicated by differential pressure values reported by the MHP falling below −100 Pa.

• The 40 s moving average of a 20 s moving-window variance of the MHP-derived TAS is above 5 m² s⁻².

• The flight_flag is 0.

Data users should be aware that extending multihole probe calibration coefficients beyond angles tested in the wind tunnel can result in highly nonlinear errors in the estimation of the angle of attack and the sideslip angle. Such errors significantly impact wind estimation. As such, users are advised to only use wind data where the wind_flag variable is equal to 0. These values are well within the calibrated range of the multihole probe.

Because the variables associated with the POPS dataset might be challenging for some data users to interpret and understand, here we provide an overview of the variables associated with this sensor. To calculate flow-corrected particle concentrations from the POPS data, the number of particles counted in each bin is divided by the measured flow rate in the POPS sensor multiplied by a ratio of the ambient temperature to the sensor temperature:

where Nax is the flow-corrected particle concentration for bin x, Xax is the POPS particle count for bin x, F_POPS is the flow rate measured for POPS, T_a is the temperature of the ambient air, and T_POPS is the laminar flow element temperature (POPS_Temperature in the dataset). This flow-corrected variable (POPS_partconc for the total aerosol concentration or POPS_binX_partconc for bin-resolved particle concentrations) is the recommended variable for determining the aerosol number concentration from the RAAVEN. Additionally, the NetCDF files provide information on the standard deviation of POPS measurements over 1 s (POPS_STD), the pressure inside POPS (POPS_Pressure), the instrument flow rate (POPS_Flow), the laser diode monitor and temperature (POPS_LDM and POPS_LDtemp), the combined total particle count (POPS_HistSUm), and a flag for the implementation of a manual binning process (POPS_useman). To allow users to plot size distributions, there is a variable that describes the edges of the 24 bins used by POPS (POPS_Bin_Edges). There is also a flag included in the TRACER-UAS data stream for the POPS aerosol spectrometer. This flag is based on different values for the ones, tens, hundreds, and thousands places. The ones digit is set to 0 if data are OK and 1 if either the aircraft is not in flight or the inlet filter is suspected to be in place. The tens digit reacts to the temperature of the sensor. POPS is designed to function optimally at temperatures below 45 °C, and increasing temperatures impact the measurement uncertainty. Therefore, the flag is set to 0 if the temperature is less than 45 °C, 1 if the temperature is between 45 and 48 °C (uncertainty <3 %), 2 if the temperature is between 48 and 50 ° C (uncertainty <7 %), and 3 if the temperature is greater than 50 ° C (do not use, uncertainty too high). The hundreds place is set based on the observed standard deviation of the measurement, with lower standard deviations being assigned 0 (good data) and standard deviations above 14 being assigned 1. Under higher standard deviations, the user is advised to use the first two bins with caution, as uncertainty at the smaller size ranges can be up to 35 % in this case. Finally, the thousands place is assigned based on the flow rate of the airstream being sampled. This flag is set to 0 for good data and is set to 1 when the flow rate is lower than 2 cm³ s⁻¹, as the lower flow rate increases uncertainty in the measured quantities.

Finally, there are two additional flags included in the RAAVEN data files to allow data users to easily understand the aircraft's flight state and support selective sampling during specific flight regimes. These flags are the “Flight_Flag” introduced earlier in the paper as well as a second “Flight_State” flag, which is a three-symbol binary variable. The Flight_State flag offers insight into whether the RAAVEN is flying straight (0 in the ones place) or is turning (1 in the ones place), whether the RAAVEN is descending (0 in the tens place), level (1 in the tens place), or ascending (2 in the tens place), or whether the RAAVEN is in flight (1 in the hundreds place) or not (0 in the hundreds place). For example, if a data user wanted to analyze straight, level flight legs, they would search for data with a Flight_State equal to 110.

The accuracy of the RAAVEN observations has been evaluated in previous studies. This includes a comparison of RAAVEN data with measurements collected by radiosondes launched from the Barbados Cloud Observatory (de Boer et al., 2022) and comparisons supported by radiosonde and tower data collected by the US DOE ARM SGP facility (de Boer et al., 2024).

4.2 University of Oklahoma CopterSonde

Raw data from the CopterSonde are stored on an SD card as binary files and then converted to a0-level NetCDF. Subsequently, data go through a process of averaging, filtering, and objective quality analysis to optimize the quality of observations. Since data from the Pixhawk are logged at a faster rate of 20 Hz than the temperature and humidity sampling rate of 10 Hz, the position and rotation data gathered by the Pixhawk are downsampled to 10 Hz to ensure a standard timeline of observations. After achieving a common time coordinate, offsets determined in the Oklahoma Climatological Survey calibration chamber are applied to each sensor. Every CopterSonde is calibrated prior to deployment, so each one has unique offsets due to minor differences in sensors. To eliminate spurious, high-frequency signals in data, the attitude data (roll, pitch, and yaw), temperature, and relative humidity data have a low-pass finite-impulse response (FIR) filter applied, described in Greene et al. (2022). Sets of three identical temperature and humidity sensors were used to ensure agreement between observations. Acceptable thresholds for sensor bias and standard deviation were experimentally determined during sensor calibration and characterization studies; if an individual sensor exceeds those thresholds, the sensor's observations are removed. Then, all remaining thermistors and humidity capacitors are averaged, and the data are binned to 5 m. This combination of sampling rate and vertical resolution ensures at least 16 observations per sensor in each bin.

Wind direction is estimated during the flight by altering the yaw angle to minimize the roll to optimize stability and promote flow into the sensor scoop. A wind speed estimate comes from the pitch angle, and in cases of high wind, the UAS will automatically return home to avoid battery fatigue or failure. In postprocessing, a more accurate horizontal wind vector is derived using a more robust linear model on the roll, pitch, and yaw while accounting for the aircraft geometry. In cases of very low wind speeds, the autopilot struggles to calculate the true wind direction, and for wind speeds less than 2 m s⁻¹ the wind direction values are considered questionable. In data comparison calculations, the wind speed and direction were removed when the wind speed was less than 2 m s⁻¹ (Table 3).

Table 3Data comparison statistics from 4425 data points of co-located vertical profiles from the RAAVEN and CopterSonde (CS).

^a Sensor uncertainties provided by the manufacturer. Values with an ^b mean biases based on comparisons to radiosondes reproduced from de Boer et al. (2024).

Download Print Version | Download XLSX