University of Colorado and Black Swift Technologies RPAS-based measurements of the lower atmosphere during LAPSE-RATE

. Between 14 and 20 July 2018, small remotely-piloted aircraft systems (RPAS) were deployed to the San Luis Valley of Colorado (USA) together with a variety of surface-based remote and in-situ sensors, and radiosonde systems as part of the Lower Atmospheric Profiling Studies at Elevation – a Remotely-piloted Aircraft Team 25 Experiment (LAPSE-RATE). The observations from LAPSE-RATE were aimed at improving our understanding of boundary layer structure, cloud and aerosol properties and surface-atmosphere exchange, and provide detailed information to support model evaluation and improvement work. The current manuscript describes the observations obtained using four different types of RPAS deployed by the University of Colorado Boulder and Black Swift Technologies. These included the DataHawk2, the Talon and the TTwistor (U. of Colorado) and the S1 (Black Swift 30 Technologies). Together, these aircraft collected over 30 hours of data throughout the northern half of the San Luis Valley, sampling altitudes between the surface and 914 m AGL. Data from these platforms are publicly available through the Zenodo archive,


Introduction
During the summer of 2018, the University of Colorado, along with the Earth System Research Laboratory (ESRL) of the National Oceanic and Atmospheric Administration (NOAA) and National Center for Atmospheric Research (NCAR) hosted the annual meeting for the International Society for Atmospheric Research using Remotely-piloted Aircraft (ISARRA;de Boer et al., 2019a). Along with this conference the organizers set up a "flight week", where participating groups could conduct remotely-piloted aircraft system (RPAS) flights in a coordinated manner to make scientifically relevant measurements of the lower atmosphere. This campaign, titled "Lower Atmospheric Profiling Studies at Elevation -a Remotely piloted Aircraft Team Experiment" (LAPSE-RATE, de Boer et al., 2020a;2020b) brought together a variety of teams to Colorado's San Luis Valley for a week of atmospheric science-centric RPAS operation. A summary of the LAPSE-RATE campaign, including an overview of meteorological conditions and the San Luis Valley sampling environment, is provided in an overview paper at the beginning of this special issue (de  (Koch et al., 2018;Frew et al., 2012;Houston et al., 2012), extensive operation of RPAS at high latitudes (de Boer 2019b;de Boer et al., 2019c;de Boer et al., 2018;de Boer et al., 2016;Cassano et al., 2016;Cassano, 2014;Intrieri et al., 2014;Knuth et al., 2013) and the use of these platforms to understand turbulence around the world, including in tropical cyclones (Cione et al., 2019). Similarly, BST has an extensive background of conducting RPAS operations for environmental science as evidenced by a variety of studies and synthesis papers (Elston et al., 2015; temperature through a calibrated semiconductor, and relative humidity using a capacitive sensor. To measure surface and sky brightness temperatures, DataHawk2s are equipped with up-and downward-looking thermopile sensors 115 (Semitec 10TP583T with custom electronics). To provide additional information on thermodynamic structure, DataHawk2s have also carried the commercially-available iMet1 radiosonde package, providing comparative information on position (GPS), temperature (bead thermistor), pressure (piezoresistive) and relative humidity (capacitive), as well as the Vaisala RSS421 sensor that is similar to the sensor suite used for the popular RS-41 radiosonde. However, neither the iMet1 nor RSS421 were installed during LAPSE-RATE. Finally, to estimate winds 120 using the DataHawk2, information from the onboard GPS, Pitot tube and inertial measurement unit (IMU) are integrated to calculate all three wind vectors. This calculation is completed using equations presented in van den Kroonenberg (2008), under the assumption that the aircraft instantly reacts to perturbations in the local wind vector, meaning that the sideslip angle (β) is assumed to be zero and that the angle of attack (α) is set to be equal to the mean pitch of the aircraft in level flight. These sensors are logged at a variety of different acquisition rates. Specifically, 125 the finewire measurements were logged on board the aircraft at 800 Hz, while other variables were logged at 100 Hz.
All data are interpolated to a common 10 Hz clock in the archived LAPSE-RATE datasets.
In addition to the DataHawk2, CU collaborated with BST to operate a Black Swift S1 RPAS ( Figure 1b) during LAPSE-RATE. This platform was originally designed as an advanced, fully autonomous survey and mapping platform. Equipped with the Black Swift Technologies SwiftCore flight management and autopilot system, remote operation and flight planning is done from a tablet. Take-off is performed through non-assisted hand launches, and the advanced landing algorithm provides for robust and precise autonomous belly landings. The aircraft has a 1.7 m wingspan and a gross take-off weight of 2.5 kg. When operated with a 13,600 mAh LiPo battery, it allows for an endurance of up to 1.5 hours at a cruise speed of 17 m s -1 , depending on the mission profile and environmental 135 conditions. The aircraft has a high operational ceiling and has been used to perform mapping missions at altitudes up to 14,000 feet in Colorado. It has been employed for surveying work, land management, crop damage assessments, and large area ecological studies. During LAPSE-RATE, the S1 was equipped with the Black Swift Technologies Multihole Probe (BST MHP). This probe includes five pressure ports to measure the system's airspeed, angle of attack and sideslip angle that can be combined with information from an INS and GPS to provide estimates of the 140 horizontal and vertical wind speed and direction. The BST MHP includees an integrated E+E Elektronik EE-03 sensor for environmental humidity and temperature. This sensor has a stated accuracy of +/-3% relative humidity for environmental conditions between 10-100% RH and a temperature accuracy of +/-0.3 C at an ambient temperature of 20 C. All data collected using the S1, with the exception of the temperature and humidity were logged at 100 Hz.
The temperature and humidity values were logged at 4 Hz. The S1 collected 10.13 hours of data during LAPSE-145 RATE.
Beyond the DataHawk2 and S1, CU operated two other aircraft on a more limited basis. The first of these was the X-UAV Talon RPAS (Figure 1c). These systems are small (1.7 m, 3 kg) airframes made of EPO foam and were outfitted with the PixHawk2 autopilot system running the open source Ardupilot Plane software. For LAPSE-RATE, the Talons were set up to fly for approximately 45 minutes on a single battery at a cruise speed of 18 m s -1 . As with the DataHawk2, this relatively slow flight speed provides high resolution measurements from the custom payload that was flown on this platform. The Talons operated for LAPSE-RATE carried a Vaisala RSS904 sensor (similar sensor module to that used in the Vaisala RS-92 radiosonde) to make measurements of temperature, pressure and humidity.
The RSS904 features a capacitive wire temperature sensor with a 0.1 C resolution, a thin-film capacitor humidity sensor with a resolution of 1 %, and a silicon pressure sensor with a measurement resolution of 0.1 hPa, and expected 155 accuracies of 0.2 C, 2% and 0.4 hPa, respectively, in temperature, RH and pressure. The Talon collected 4.2 hours of data during LAPSE-RATE.
The RSS904 sensor suite was integrated into a custom 3D-printed nosecone. In addition to measurements from the RSS904, wind speed and direction were estimated by the autopilot system, using a combination of the GPS velocities, aircraft attitude measurements from the inertial navigation system and aircraft airspeed. Aircraft navigation data were 160 logged at 10 Hz, while the pressure, temperature, and humidity values from the RSS904 were logged at 2 Hz. For a limited number of Talon flights, the aircraft carried a "microsonde" custom temperature, pressure and humidity sensor suite developed at CU. The microsonde was developed to be integrated into a Lagrangian drifter called the Driftersonde (Swenson et al., 2019) that could be launched from small RPAS for atmospheric research. The microsonde includes a MS8607 PTH sensor from TE Connectivity, which consists of a piezoresistive pressure sensing 165 element, which measures both barometric pressure and temperature. The piezoresistive Micro-Electro-Mechanical Systems (MEMS) measure atmospheric pressure relative to a vacuum inside the MEMS, sealed by a thin membrane.
A Wheatstone bridge logs temperature by measuring the temperature-dependent resistance. Additionally, the MS8607 includes a capacitive sensing element that features a dielectric polymer film, sensitive to humidity, between two electrodes to measure relative humidity. In addition to the MS8607, the microsonde includes a CAM-M8Q GPS 170 module (ublox) that provides reception of up to 3 global navigation satellite systems (GNSS) (GPS, Galileo, GLONASS, BeiDou) at a given time. The integration of the microsonde on the Talon during LAPSE-RATE was, in part, to provide a dataset to allow for direct comparison between the measurements it produced to those from the RSS904, which is considered to be an industry standard sensor suite. Microsonde data were logged at 1 Hz.

175
Finally, the other aircraft operated by CU that saw limited action during LAPSE-RATE was the TTwistor RPAS were logged at 100 Hz, and the RSS904 variables were logged at 2 Hz.
The extended endurance of the TTwistor, along with advanced implementation of autopilot capabilities and a surface tracking vehicle (Figure 1d) allows this platform to be deployed in a "follow-me" mode. In this mode of operation, the flight team is inside of a moving vehicle and the aircraft is programmed to stay within a given distance of the 195 vehicle while in flight. This allows for the execution of extended horizontal transects for collection of data on spatial variability and gradients. Follow-me has been used to make measurements of outflow boundaries associated with severe storms, and during LAPSE-RATE was implemented to attempt to better understand the influence of surface heterogeneity on the overlying atmospheric state. The TTwistor collected 7.49 hours of data during LAPSE-RATE.  Figure 3). This site featured large crop circles, each having different states of irrigation and plant activity. At this site, two primary flight modes were employed. The first of these was a basic profiling mode, where the aircraft conducted repeated vertical ascents and descents while following an orbital flight track. These flights were designed to evaluate the development of the boundary layer and temporal evolution of its structure during the morning hours. Profiles typically ranged 215 between the surface and 500 m above ground level (AGL). In addition to these profiling flights, additional emphasis was placed on attempting to measure differences in boundary layer development over different surface types, mainly between crop circles with different amounts of plant coverage and irrigation levels. For those flights, the aircraft were operated in a loiter pattern over two different crop circles for extended time periods.

220
With the exception of 19 July, the BST S1 was operated at the mouth of the Penitente Canyon, just to the north of the intersection between routes 38A and 40G ( First, the S1 shows an influence of heating of the temperature sensors while the aircraft is on the ground. Because of 265 this heating, the S1 temperature sensor provides excessive temperature readings for the first portion of the flight, resulting in super-adiabatic lapse rates extending from the surface to around 100 m altitude. This is resolved after the aircraft has been in the air for several seconds. The DataHawk2 slow temperature sensor shows similar structures, but their prevalence is greatly reduced if analyzing the coldwire (fast) temperature measurement. Notably, both of these platforms have some variability near the surface that is likely to be a result of measurement error there, rather than 270 atmospheric features near the surface. In addition to these near-surface features, it is also notable that the TTwistor has more variability in its measured wind speed than the other platforms. Data shown in Figure 7 do not include those collected during manual flight of the aircraft, eliminating the possibility of this factor impacting the estimated winds.
The TTwistor was flown somewhat later in the day during the 18 th of July, and traversed a significant area during its "follow-me" flight operations. During these afternoon flights, the aircraft passed through several areas of gusty winds.

275
While the exact cause for these winds is not determined, we speculate that they were associated with outflows from large convective eddies and the significant thermal gradients established in the valley due to the variability of solar exposure and surface land cover. The latter variability is noticeable in the specific humidity data shown in Figure 7 as well.

285
As discussed in de Boer et al. (2020b), LAPSE-RATE data files are provided in NetCDF format with a common file name structure. This common structure (xxx.ppppp.lv.yyyymmdd.hhmmss.cdf) provides information on the institute generating the data (xxx), the platform used to collect the data (ppppp), the level of data processing (lv), and the date and time that the measurements were obtained. The institutions covered in the current paper include the University of Colorado Boulder (UCB) and Black Swift Technologies (BST), with platform identifiers including the DataHawk 290 2 (DATHK), the S1 (BSTS1), the Talon (TALN3) and the TTwistor (TWSTR).
The teams provided different levels of processing and quality control for the measurements from these platforms.
Processing of the DataHawk2, Talon and TTwistor resulted in files at the b1 level. The S1 files were only processed to the a1 level. Similar limits were applied to all three UCB datasets to provide a b1 level dataset. These values were 295 selected as physical limits that should not be exceeded for data collected during LAPSE-RATE, and are reviewed in Table 2. As a reminder, the levels that were defined for LAPSE-RATE in general included:

a0:
Raw data converted to netCDF a1: Calibration factors applied and converted to geophysical units b1: QC checks applied to measurements to ensure that they are "in bounds". Missing data points or those with 300 bad values should be set to -9999.9

c1:
Derived or calculated value-added data product (VAP) using one or more measured or modeled data (a0 to c1) as input Processing of the DataHawk data included calibration of the coldwire temperature and calculation of the winds. The 305 coldwire voltages were calibrated against temperatures derived from the SHT sensor. This relationship is expected to be linear and the calibration relationship is derived using a first-order polynomial fit. This fit is calculated on a flightby-flight basis, leveraging temperature measurements from parts of the dataset that were collected while the aircraft was in flight. In addition, the first 1000 datapoints (100 s at 10 Hz) are also omitted from this calculation to avoid contamination from the surface heating of the temperature sensor discussed earlier. DataHawk2 winds are calculated 310 by leveraging observations from the pitot tube, the GPS, and the onboard inertial measurement unit (IMU).
Leveraging information from these sensors, the autopilot computes wind in inertial coordinates as the vector difference of the velocity of the vehicle relative to the ground, minus the velocity of the vehicle relative to the air. The later vector is estimated in magnitude as the airspeed from the pitot-static sensor, and in direction as the orientation of the vehicle longitudinal axis in inertial coordinates. This assumes that the vehicle is pointed into the relative wind (i.e.

315
with only small deviations in sideslip and angle of attack), which is thought to be a good assumption for wind variations with timescales longer than about 1 sec for a vehicle of this scale. Because the DataHawk2 winds calculated from these sensors include a periodic signal that corresponds to the frequency of an orbital circle, additional filtering is applied to these winds. This technique applies a 15 th order low pass infinite impulse response (IIR) filter to the originally calculated winds to produce a smoothed version. Additionally, using yaw data from the autopilot, the mean 320 wind is calculated over any single orbit. Finally, the high-frequency component of the originally calculated winds (difference between the calculated wind and the output from the lowpass filter) is added back to the wind calculated across the orbital means. As a result, these wind estimates are meant to primarily provide perspective on the mean wind structure with altitude, in addition to the higher frequency turbulent structure.

325
The Talon and TTwistor data also require the calculation of wind components. For both datasets, winds are calculated using formulae available in van den Kroonenberg et al. (2008). This technique takes input from the onboard GPS, IMU and airspeed sensor to calculate three-dimensional wind components. The TTwistor carried a multihole pressure probe to measure angle of attack and sideslip angles, while the Talon only carried a standard Pitot tube for estimating airspeed. As a result the Talon angle of attack and sideslip were estimated to be a constant 1.75 degrees and 0 degrees, 330 respectively, for the entire flight. This assumption results in the omission of fine turbulent motion that would not be captured without understanding of the angle of airflow over the airspeed sensor. For both platforms, a series of corrections are applied to account for offsets in the calculated true airspeed (TAS), and potential angular offsets between the IMU and the airspeed sensor (yaw and pitch offsets). These offsets are calculated in an iterative manner by minimization of the variance of the calculated winds with an order of TAS, pitch, yaw, TAS and yaw. The last 335 two TAS and yaw corrections are applied to apply a finer scale correction than possible in the first round.

Data Availability
The data files from the LAPSE-RATE project are generally being archived under a LAPSE-RATE community 340 established at the Zenodo data archive (https://zenodo.org/communities/lapse-rate/). From here, LAPSE-RATE observations are available for public download and use. Data Object Identifiers (DOIs) were automatically generated by the Zenodo archive at the data version and product level. Data from the different sources described above are posted as individual datastreams on the archive, with each of the platforms described in the previous section having their own DOI. It is important to note that each platform may have several different levels of data available. Therefore, 345 data products with different levels of processing and quality control may be provided with separate DOIs. This means the files and data described in this publication are spread across a variety of DOIs, and that additional DOIs could be created in the future that include LAPSE-RATE data, as additional data products are developed.
As of the writing of this manuscript, the CU DataHawk2 dataset (de Boer et al., 2020c)  As with the Talon data, there are multiple versions, and the most current version as of the writing of this manuscript is 3.0. Finally, the BST S1 datasets (Elston and Stachura, 2020)    -120<lon<-100 N/A -120<lon<-100 -120<lon<-100 u, v (m s -1 ) -50<u,v<50 N/A -50<u,v<50 -50<u,v<50 w (m s -1 ) -20<w<20 N/A -20<w<20 -20<w<20 Wind Speed (m s -1 ) 0<wspd<100 N/A 0<wspd<100 0<wspd<100 Table 2: Quality control limits applied to the different data b1 datasets. To date, S1 data are only available as a1 DataHawk2, the BST S1, the CU Talon, and the CU TTwistor mounted on top of a tracker vehicle for launch prior to "follow me" operations.      Profiles of (left to right) temperature, specific humidity and wind speed from the DataHawk2, S1, Talon and TTwistor (top to bottom). Colors indicate the date of flight. Please note that the altitudes are shown as "above ground level" (AGL), and some aircraft were operating at different locations each day and from one another, meaning that the ground levels in this figure represent different elevations relative to sea level.