The RAAVEN UAS (Fig. 1) is a fixed-wing UAS with a wingspan of 2.3 m that has been operated by the University of Colorado Boulder since 2019. The RAAVEN's airframe is based on a custom-manufactured model from RiteWing RC. The airframe has been updated to meet the needs of atmospheric science missions spanning a variety of environments. The RAAVEN leverages the PixHawk2 autopilot system and employs an 8S 21 000 mAh lithium ion (Li-Ion) battery pack to offer flight times around 2.5 h, depending on conditions and executed flight patterns. Specific modifications to the airframe include the integration of a tail boom to enhance longitudinal stability and improve the platform's performance. The aircraft has a top airspeed of approximately 130 km h-1, though operations during WiscoDISCO-21 were almost exclusively conducted in the 60–70 km h-1 range.

For the WiscoDISCO-21 campaign, the RAAVEN was equipped with an instrument suite derived 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. In this configuration, the aircraft was set up to measure atmospheric and surface properties to support evaluation of thermodynamic state, kinematic state, and turbulent fluxes of heat and momentum. This involves a suite of core instrumentation (see Fig. 3), including a multihole pressure probe (MHP) from Black Swift Technologies, LLC (BST); a pair of RSS421 PTH (pressure, temperature, humidity) sensors from Vaisala, Inc.; a custom finewire array, developed and manufactured at the University of Colorado Boulder; a pair of Melexis MLX90614 IR thermometers; and a VectorNav VN-300 inertial navigation system (INS). This sensor suite is logged using a custom-designed FlexLogger data logging system.

The Vaisala RSS421 sensors are identical to those used in the Vaisala RD41 dropsonde and very similar to the Vaisala RS41 radiosonde. This unit employs a linear resistive platinum temperature sensor with a resolution of 0.01 ∘C, repeatability of 0.1 ∘C, and response time (as measured within the RS41 radiosonde) of 0.5 s at 1000 hPa when moving at 6 m s-1. For relative humidity (RH), the RSS421 includes a thin-film capacitor with a resolution of 0.1 % RH and a repeatability of 2 % RH, with a temperature-dependent response time of better than 0.3 s at 20 ∘C (again, as measured within the RS41, with 6 m s-1 airflow at 1000 hPa). Finally, the pressure sensor is capacitive with a silicon diaphragm, having a resolution of 0.01 hPa and a repeatability of 0.4 hPa. For WiscoDISCO-21, a pair of these sensor modules was mounted to the top of the RAAVEN's fuselage, between the nose and the tail of the aircraft on the port side. The sensor mounting angles were offset to ensure that the two sensors would have different amounts of solar exposure as the aircraft maneuvers through the atmosphere and to allow for the detection of solar heating effects since no shading is used. Additional information on atmospheric thermodynamic state is available from an E+E EE-03 sensor that is integrated into the BST MHP and from a Sensiron SHT-85 sensor that is integrated in the custom finewire array. The EE-03 has a temperature accuracy (at 20 ∘C) of 0.3 ∘C, while the humidity accuracy is stated to be 3 % RH at 21 ∘C. The SHT-85 has a stated temperature accuracy of 0.1 ∘C (from 20–50 ∘C) and a repeatability of 0.08 ∘C, while the humidity sensor has a stated accuracy of 1.5 % RH and a repeatability of 0.15 % RH. Both the EE03 and the SHT-85 sensors have slower response times than the RSS421 sensor described above and are typically not used for scientific purposes unless there is a complete failure of the RSS421.

In addition to the SHT-85 sensor, the finewire array consists of two 5 µm diameter platinum wires extending over a 2 mm length, suspended in the free stream by supporting prongs. One wire is operated as a hotwire anemometer, with approximately 100 ∘C overheating compared to the ambient environmental temperature. The other wire is operated as a coldwire thermometer, with approximately 1 ∘C overheating relative to the surrounding environment. The wires have thermal time constants of 0.5 ms in a 15 m s-1 airflow regime and support a sampling frequency of up to 800 Hz to support measurement of turbulent fluctuations in velocity and temperature. An electronics module converts resistance change in the wires due to velocity or temperature variability to amplified voltages. For WiscoDISCO-21, the finewire was logged at 250 Hz by the FlexLogger, which is equivalent to a 7.2 cm minimum length scale at the RAAVEN's typical cruise airspeed of 18 m s-1​​​​​​​. Time series of these recorded data are processed during post-flight analysis to transform the voltages recorded by the fine-wire module to velocity and temperature. Additionally, these measured quantities can be fit to inertial sub-range turbulence models to wavenumber spectra over suitable time intervals, producing turbulence intensity parameters epsilon (kinetic energy dissipation rate) and CT2 (temperature structure constant). The resolution (noise floor) of these parameterizations is 2.0 × 10-7 W kg-1 for epsilon and 4.5 × 10-6 K2 m-2/3​​​​​​​ for CT2. Resolution of the raw time series is 8.3 × 10-5 m s-1 for the hotwire and 1.3 × 10-4 K for the coldwire.

In addition to the EE-03 PTH measurements, the BST five-hole probe supports measurement of airspeed, angle of attack (α), and sideslip angle (β). These measurements are combined with the GPS-based ground velocities and aircraft altitude from the VectorNav VN-300 to derive the three components of the inertial wind (u, v, w), as discussed in Sect. 4. The VN-300 can be configured in a dual-Global Navigation Satellite System (GNSS) mode, under which the relative positions of two GNSS antennae are used to calculate the platform yaw. However, this setting was not used during the WiscoDISCO-21 deployment. Under dynamic conditions, the system has a stated accuracy of 0.3∘ in GPS compass heading, 0.1∘ in pitch and roll, 2.5 m horizontal position accuracy, 2.5 m vertical position accuracy when integrating information from the barometric pressure sensor, and 0.05 m s-1 accuracy in inertial velocity. Input from the system's gyroscope, accelerometer, GNSS receiver, magnetometer, and pressure sensor is filtered through an extended Kalman filter (EKF) to produce a navigation solution. VN-300 data were logged at 50 Hz resolution during WiscoDISCO-21.

Finally, RAAVEN deploys two Melexis MLX90614 IR thermometers: one looking up from the top of the aircraft and one looking down towards the surface in level flight. These sensors are factory calibrated to work in operational temperatures between -40 and 125 ∘C and to measure target brightness temperatures between -70 and 380 ∘C. They have a high accuracy (0.5 ∘C) and a measurement resolution of 0.02 ∘C. The RAAVEN-mounted MLX90614s are not stabilized to maintain a vertical orientation, meaning that the observed target is perpendicular to the reference frame of the aircraft. This requires some care when interpreting measurement from time periods when the aircraft is conducting pitch or rolling maneuvers. For WiscoDISCO-21, we leveraged the “I” version of this sensor, which has a 5∘ field of view. These sensors have a broad passband range of 5–14 µm, meaning that while it covers the infrared atmospheric window, it is also subject to radiation emitted by water vapor and other radiatively active gases. This means that there is a significant depth of atmosphere between the aircraft and a given target (e.g., cloud, surface), and atmospheric gases influence the temperature reading. Despite this range spanning the 9.6 µm O3 band, the relative proximity of the sensor to targets of interest (e.g., surface, clouds) means that this overlap is not expected to significantly influence the readings, due to the integrated path length being relatively small. Therefore, if absolute accuracy of brightness temperature is important, the sensor should be operated in close proximity to a target of interest. However, relative contributions from different surface types or atmospheric conditions can still be easily distinguished despite a lack of absolute calibration for extended distance sensing. Such gradient detection can be useful for detecting surface inhomogeneities, or for understanding whether the aircraft is operating under cloud or clear sky.

The DJI M210 quadcopter was equipped with a 3-D printed top-mounted bracket for holding a 2B Technologies personal ozone monitor (POM) and an Intermet Systems iMET-XQ2 meteorology sensor (Fig. 2). The copter had a ∼ 15 min flight time with the onboard sensors without a camera. The POM measures ambient ozone using UV absorption and active humidity subtraction by measuring a whole-air sample and an ozone scrubbed sample in a 10 s duty cycle. The POM records data to its internal data storage at 10 s intervals with a log number and timestamp along with GPS coordinates and instrumentation metrics (optical cell pressure and temperature). The iMET system records temperature, humidity, and pressure along with GPS coordinates and a timestamp to internal data storage. Each instrument (the POM and iMET) had individual data logging systems and separate power supplies. Both the POM and the iMET had GPS capabilities with the POM logging inconsistently and the iMET logging GPS more consistently. Each instrument and the UAS flight recorder logged timestamps. The iMET recorded observations of temperature, relative humidity, humidity temperature, and pressure at a frequency of 10 Hz. The POM recorded ozone observations at a frequency of 0.1 Hz. The POM, iMET, and M210 timestamps drifted up to 60 s from the other logged data. The flight log recorded the M210 positioning (altitude, latitude, longitude) at 100 Hz. The M210 flight logs, iMET data, and POM data were each downloaded separately after each series of flights.

A Halo Photonics Stream Line XR Doppler lidar (Pearson et al., 2009) was deployed on the roof of the Chiwaukee Prairie air monitoring station (Fig. 3), approximately 3 m a.g.l. This is the same system that is regularly deployed as part of the Space Science and Engineering Center (SSEC) Portable Atmospheric Research Center, SPARC (Wagner et al., 2019). The Doppler lidar actively emits pulses of near-infrared radiation at a wavelength of 1.5 µm. This wavelength is long enough that molecular scattering causes little attenuation of the signal, but it is short enough that it is sensitive to aerosols that are suspended within the planetary boundary layer.

The Doppler lidar uses velocity-azimuth display (VAD) scans of the Doppler lidar to retrieve profiles of wind speed and direction. In VAD, an instrument capable of measuring along-beam velocity (like a Doppler radar or lidar) stares at multiple azimuths at a non-zenith elevation angle over a short period of time and then reconstructs the profile of winds above the lidar by assessing how the along-beam velocity changes as a function of azimuth and range. For WiscoDISCO-21, the VAD scans were configured with six azimuthal stares per profile (at azimuths of 0, 60, 120∘, and so on) with an elevation angle of 70∘. Range gates were 18 m. VAD scans were conducted every 4 min, and each VAD took approximately 45 s to complete. Between VADs, the lidar reverted to vertical stares in order to capture profiles of backscatter and vertical velocity. Since the lidar depends on the presence of scatterers to have a detectable signal return, the depth of the retrieved wind profiles varied significantly throughout the experiment from as shallow as 200 m to as deep as 2 km.

Data collected by the different sensors carried by the RAAVEN during WiscoDISCO-21 were logged at a variety of different logging rates. The finewire system was logged at 250 Hz, the fastest rate of all of the sensors. The BST MHP was logged at 100 Hz and the VectorNav VN-300 at 50 Hz, the Melexis IR sensors and variables related to finewire status were logged at 20 Hz, while data collected from the PixHawk autopilot and Vaisala RSS421 sensors were logged at 5 Hz. Each logging event carried out by the FlexLogger includes a sample time from the logger CPU clock, allowing for post-collection time alignment between the different sensors. These sample times, along with artificial 5, 20, 50, 100, and 250 Hz clocks spanning the sample times between initial GPS lock and the last recorded sample time for the VN-300, are used to align the different variables to a set of common clocks, primarily through one-dimensional linear interpolation. One exception to the linear interpolation is the yaw estimate, which is circular in nature (ranging between -180 and 180∘) and therefore uses a “nearest” interpolation to ensure that the transition from 360 to 0∘ is not represented as 180. During this interpolation process, a limited number of points sharing a common sample time with another point are removed from the record. Once these time variables are established, a base_time variable is established using the 250 Hz timestamp and offsets from base_time are then calculated for all different logging resolutions.

The resampled (in time) dataset includes a variety of derived and measured quantities. Aircraft positions, including latitude, longitude, and altitude, are measured by the VN-300. The 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. Pressure altitude is subject to drift over the duration of a single flight due to atmospheric evolution over a 2.5 h window, potentially resulting in values at landing that are higher or lower than those at take-off. Similarly, the accuracy of the GPS altitude is insufficient to capture the vertical position of the aircraft to the level of detail required. To calculate a true altitude, a combination of the autopilot altitude, VN-300 altitude, and VN-300 pressure is used. First, a flight_flag variable is computed using airspeed and altitude information from the autopilot. Any data points with airspeed exceeding 10 m s-1 and an altitude exceeding 5 m a.g.l. is flagged as a time when the aircraft is flying (flight_flag = 1). The point at 200 samples (4 s) prior to the first point in the record where the aircraft is deemed to be flying is recorded as the initial take-off index, while the data point at 200 samples (4 s) after the last point in the record where the aircraft is deemed to be flying is recorded as the landing index. The difference between the autopilot altitude at these two indices is added into the flight record on a time step-by-time step basis, to correct for temporal drift in pressure. A linear fit is then calculated to relate the VN-300 pressure and the difference between the VN-300-reported altitude and the autopilot-reported altitude. This pressure-dependent altitude correction is then applied to the VN-300-reported altitude to derive a final altitude.

Wind estimation from fixed-wing aircraft requires the combination of several different measurements related to airspeed, aircraft motion, and airflow over the aircraft (see van den Kroonenberg et al., 2008). These measurements need to be of sufficient quality, and angular offsets and logging delays need to be considered and removed. For RAAVEN, true airspeed (TAS) biases have a large impact on derivation of wind speed, while the angular offsets between the MHP and INS and time lag between the GPS and in situ measurements have smaller impacts. These potential sources of error are corrected for using an optimization technique, where small adjustments are made to the individual parameters, and the combination that results in the wind solution with the smallest overall variance is selected as the correct winds.

For the RAAVEN WiscoDISCO-21 dataset, TAS is calculated using measurements from the MHP and RSS421 probe using Eq. (1) from Brown et al. (1983): 1TASi=2q‾ρ, where ρ is the density of air calculated using the static pressure reported from the MHP, temperature from the RSS421, and the specific gas constant for dry air. q‾ is defined as 2q‾=p01-94sin⁡2θa, where sin⁡2θa is the total aerodynamic angle of the MHP, calculated using the angle of attack (α) and sideslip angle (β) reported by the MHP.

Based on testing in a temperature chamber, the pressure sensors used in this version of the MHP were found to have non-linear temperature dependencies. This requires the application of an additional temperature-dependent correction to ensure that an artificial alteration of TAS with altitude was not present. Additional information on the calculation of airspeed and other quantities from the MHP can be found in de Boer et al. (2022a).

Derivation of the RAAVEN's thermodynamic measurements included multiple processing steps. First, data from the two RSS421 sensors are averaged to attempt to reduce the influence of any solar exposure of the sensors. Previous evaluations of the potential for solar contamination have not revealed any specific biases on the observation (see de Boer et al., 2022a). Over the course of the WiscoDISCO-21 campaign, the two sensors varied by less than 0.5 ∘C (Fig. 8). The averaged temperature time series was then used to calibrate the coldwire data by applying a linear fit to the relationship between the coldwire voltage and the temperature measured by the RSS421 sensor. The RSS421 relative humidity values were also averaged. Typically, the RH measurements agreed to within 2 %.

All quantities measured by the RAAVEN have data quality flags associated with them. For the RSS421-derived temperature, the flag is set to zero for good data and set to 1 for times when any of the following occur: (a) the absolute value of the difference between the temperature from either individual sensor and the output temperature is greater than 0.5 ∘C, (b) the absolute value of the difference between the output temperature and the temperature from the EE-03 sensor on the MHP exceeds 5 ∘C, (c) the recorded error flag of either RSS421 sensor is active, or (d) the aircraft is not flying. For the RH measurement from the RSS421, a similar set of criteria are used to activate the data quality flag, except the limits are set to be 6 % between RSS421 sensors and 15 % between the output RH value and the MHP-provided RH value. The relative humidity values from the MHP are significantly impacted by the exposure of that sensor to sunlight and the associated impact on sensor temperature. This is not corrected for, resulting in large fluctuations in the RH values at times. As a result, this measurement (from the MHP) only provides a reality check to ensure that the RSS421 sensors are reporting accurate values, and therefore such a large offset (15 %) is allowed. The more important comparison is between the two RSS421 sensors, which should agree much more closely, as they are the same sensor type and are mounted within close proximity of one another. The coldwire temperature data quality flag is activated when the difference between the coldwire temperature and either of the RSS421 temperatures exceeds 1 ∘C, when the absolute value of the difference between the coldwire temperature and the MHP temperature exceeds 3 ∘C, when the coldwire voltage is observed to fall outside of the 0–4 V analog range, or when the aircraft is not flying. Finally, the pressure quality control flag for the pressure measurement from the VN-300 is activated if the absolute value of the difference between the reported VN-300 static pressure and that measured by either of the RSS421 sensors exceeds 2.5 hPa. The RSS421 pressure measurements are not used because they are believed to be biased low due to the airflow passing over their location on the aircraft.

In addition to the flags discussed above, we include a three-stage flag for the wind measurements, which is set to 0 (good data), 1 (suspect data), or 2 (bad data). Data are determined to be bad if any of the following conditions were met.

The measured angle of attack or sideslip exceeds 20∘, with values between 10–20∘ flagged as “suspect”.

Finally, we included two additional flags in the data stream to allow data users to better understand aircraft flight state and support sampling during specific phases of flight. These flags include the “Flight_Flag” introduced previously, as well as a “Flight_State” flag. The Flight_State flag includes information on whether the craft is flying straight (0) or is turning (1) in the ones place, whether the aircraft is descending (0), level (1), or ascending (2) in the tens place, and whether the aircraft is in flight (1) or not (0) in the hundreds place. If, for example, a data user wanted to analyze straight, level flight legs, they would search for data with Flight_State equal to 110. These flags are derived from information from a combination of sensors, including the altitude variable described above, the aircraft yaw, and the Flight_Flag variable described earlier on in this paragraph.

The accuracy of the RAAVEN observations has been evaluated in previous studies. For example, a comparison of RAAVEN data with measurements collected by radiosondes launched from the Barbados Cloud Observatory was conducted in recent work from de Boer et al. (2022b). While radiosondes in that evaluation were launched approximately 20 km to the southeast, the air sampled by both systems was largely representative of the marine boundary layer, implying limited spatial variability. In that evaluation, the observations from the RAAVEN were very well correlated with those from the radiosondes and do not show any positive or negative bias, supporting the idea that the RAAVEN measurements provide an accurate depiction of the lower atmosphere. In addition, recent work allowed for direct comparison of RAAVEN data to observations collected by radiosondes and a 60 m tower at the US Department of Energy's Southern Great Plains (SGP) research site. That study, de Boer et al. (2022a), similarly provided confidence in the RAAVEN observations, showing close statistical agreement between the different data sources.

Data from the M210 flight controller, the POM, and the iMET were all logged to individual instrument internal data storage with independent timestamps. The average flight time of the M210 was 13.96 min. The POM instrument logged data at 0.1 Hz. The iMET logged data every 10 Hz, and the M210 flight log recorded UAS GPS positioning and flight statistics at 100 Hz. The ozone concentrations from the POM are adjusted to calibrated values, where ozone calibrations were conducted before every set of two flights for the M210 using a 2BTech model 306 ozone calibration source (Fig. 9). Data quality flags are established as 0 being no concern, 1 being time flag, and 2 being calibration and time flag. The time flag indicates flights where the time offset between the M210 and the instrument time offset is large (iMET >10 s or POM >30 s). The calibration flag indicates when the POM was not responsive to the ozone calibration source (Flight 5 on 24 May) after an over-water flight. All times were averaged to 90 s and compressed to the time window of observations for a single M210 ascent using the M210 timestamp. A timestamp for 90 s averaged data from all instrumentation on the M210 was generated by using the M210 timestamp as primary and adjusting to a time offset in either the POM or the iMET for the start of a flight; then each variable was averaged for every 90 s interval of the flight. A 1σ standard deviation is presented as the uncertainty for the 90 s averages. The iMET observations of temperature, relative humidity, pressure, and humidity temperature are presented using the 90 s averages with uncertainty as 1σ standard deviations. Each flight ascent start and end were determined by observed changes in atmospheric pressure by the iMET sensor, altitude change in the M210, or noted time of ascent in field notebook for the POM. The altitudes for each observation were obtained by averaging the M210 flight log altitude data for the 90 s timestamps. The flight data timestamps varied slightly for each data source. The POM time drift was the most pronounced, with an average difference between the iMet of ∼ 24 s. The POM's time was adjusted manually throughout the campaign as the time would drift over the course of one flight. The average difference between the iMet and the M210 over 20 flights was ∼ 4 s. Only 20 % of flights had a time difference between iMET and M210 greater than 10 s. Instrument battery loss occurred for the iMET system, which resulted in lost data for two flights​​​​​​​ on 26 May 2021.

Intercomparison between observations made via instrumentation on the M210 at 5 m a.g.l. and at the Wisconsin DNR ground station show a linear agreement between the observations (Fig. 10). The linear agreement is better for the iMET temperature and the ground station temperature with R2=0.970 in comparison to R2=0.955 for O3 observations. The O3 linear fit, O3(POM)=0.944 (±0.044)​​​​​​​ O3(DNR)-3.3 (±1.9), has a negative intercept. The uncertainties in the POM's O3 concentrations are much larger than uncertainties in the ground station instrumentation. The linear agreement between the different instrumentation on separate observation platforms demonstrates that the M210 platform instrumentation provides an accurate, albeit less precise, representation of the atmosphere.