The presented CopterSonde airframe is based on a modified version of the Lynxmotion HQuad500 construction frame kit from RobotShop Distribution Inc. The CopterSonde is a rotary-wing vehicle that has four fixed pitch rotors mounted at the end of four almost equally spaced arms attached to the main body, also known as quadcopter. The structure of the quadcopter is made of G10 fiberglass plates, carbon fiber tubes, aluminum tube-clamps, and stand-offs. A combination of T-motor U3 700 KV motors with carbon fiber T-style propellers 11′′×5.5′′ (27.94cm×13.97cm) is attached at the end of each arm. The motors are controlled by a Lumenier 30 A BLHeli32 4-in-1 speed controller which is powered by a single four-cell 6750 mAh lithium polymer battery. The battery has its own compartment inside the CopterSonde to avoid exposing it to extreme environments and to keep it within operating temperatures. Furthermore, all of the CopterSonde's electronics are covered by a shell to protect them from external hazards. The CopterSonde technical and performance specifications are summarized in Table .

The original frame was intended for simple recreational flights, in particular for long-endurance first-person view (FPV) flights. Therefore, the quadcopter had to undergo some modifications to accommodate the desired components and repurpose the functionality of the aircraft for weather data collection. Several components were fitted in the CopterSonde, which included a high-precision Here+ V2 Ublox M8P GPS that has the ability to process real-time kinematic (RTK) GPS positioning with a precision on the order of few centimeters. Also, a lidar-based rangefinder was mounted on the CopterSonde to accurately measure the height close to the ground which works in conjunction with a precision landing system based on an infrared camera. Figure  shows the 3-D-printed mount supports for the lidar device and the precision landing camera, respectively, which were integrated into the structure of the quadcopter. The 3-D-printed parts, including the shell that encloses the body of the CopterSonde, were modeled using the SolidWorks^® computer aided design (CAD) program. Several iterations of the different CAD parts were printed with polylactic acid (PLA) plastic using a LulzBot^® TAZ 6 3-D printer until a perfect fit was attained. Furthermore, the digital models were also useful to study the airflow behavior and its trajectory around such parts and the CopterSonde by using computational fluid dynamic (CFD) simulations, which is discussed in Sect. .

The arrangement of the electronic components within the CopterSonde was carefully planned to increase its modularity and facilitate performing routine maintenance. In particular, the payload was strategically placed at the front-most section of the CopterSonde. Consequently, the sensors are subjected to a cleaner airflow by keeping the payload always facing into the wind. As a result, undesired data contamination (produced by sources such as the heat emanated by the CopterSonde) is significantly reduced. This technique is better described in Sect. .

The CopterSonde is equipped with a Pixhawk Cube 2.1 autopilot board (Hex Technology, Sha Tin, Hong Kong), which is used as the main controller for flight stabilization and navigation. The Pixhawk is an open-hardware project that provides users with a readily available, economic, and high-end autopilot board solution for academic, hobby, and industrial communities. It contains a powerful microcontroller capable of executing complex autonomous missions. The Pixhawk board runs the ArduPilot autopilot code, which is a free software package that can be redistributed and modified under the terms of the GNU General Public License version 3 (GPLv3). This means that anyone is free to use all of the code and tools provided in the ArduPilot GitHub repository without authorization or involvement from the ArduPilot team. However, it is the responsibility of anyone using the code to inform the end user that open software is used and the source must be provided. Therefore, under the given license terms, the code source release made by CASS is available to the users in , which is in accordance with ArduPilot's GPLv3 license.

The motivation behind the decision taken to modify an existing open-source autopilot code was the flexibility of incorporating additional desired features not dependent on proprietary codes from companies, which usually do not follow the same research line and are prone to being discontinued. By having full access to the UAS autopilot code and being able to modify it as desired, it allowed for considerable reduction in the required electronic hardware by centralizing functions in a single processing unit. For instance, separate data loggers for the weather sensors were not needed, as the Pixhawk autopilot board has sufficient computing power to accommodate the extra data. This also came with the benefit of weight reduction which is crucial to improving flight performance and endurance. The official ArduPilot code by default does not support the desired meteorological sensors nor the adaptive flight behaviors for optimal atmospheric sample acquisition. Thus, the ArduPilot code has undergone modifications to incorporate custom user-code functions to adapt it to the specific application.

In the current version of the CASS–ArduPilot code run by the CopterSonde, a set of custom functions was added on top of the original ArduPilot code. These are itemized with their motivation and description as follows.

Code libraries to read the bead thermistors distributed by International Met Systems (iMet) and HYT-271 humidity sensors distributed by Innovative Sensor Technology (IST) were added into the autopilot code to achieve weather sensor integration. The communication between the sensors and the autopilot is done over a single bus using the I2C protocol, capable of sampling and storing up to eight sensors at 20 Hz each to an internal SD card.

The approach used to estimate the 2-D wind vector on the CopterSonde is based mainly on the tilt measurements given by its onboard inertial measurement unit (IMU). By keeping the magnitude of the roll angle of the CopterSonde around zero, a quasi-symmetric condition to the oncoming wind can be achieved, and thus the wind speed can be calculated based on the pitch angle and a function of projected area normal to the wind (discussed in Sect. ). This approach is referred as the “wind vane mode”. Concisely, the concept of the wind vane mode is to ensure that the CopterSonde is pointing headwind by changing its yaw angle in the direction that the roll is minimized.

The advantage of this approach is that there is no need for an additional dedicated airspeed sensor or anemometer, which is challenging to implement properly on a multicopter and would also reduce the available space and weight limit for other valuable payload components. However, the inherent rotor wash causes a big limitation within this approach, because it disturbs the aerodynamic field around the CopterSonde. Hence, resolving for small wind fluctuations becomes highly unreliable.

The wind vane mode algorithm has to quickly calculate the wind vector to keep the CopterSonde within the quasi-symmetric conditions while ascending and descending. Also, the algorithm does not resolve for the fine-scale wind vector in order to reduce mathematical complexity and thus reduces the computing time on the Pixhawk controller. Therefore, the kinematic equations of motion of the CopterSonde were used for the algorithm over the more complex dynamical model.

The reference frame convention used for vector calculations is shown in Fig. , where nyz is a unit vector in the direction of X normal to the plane YZ of the inertial frame I:{X,Y,Z}, θ and ϕ are pitch and roll angles, respectively, and eθ and eϕ are unit vectors of the body reference frame. The angle between the vector v=(eθ×eϕ)xy, normal to the CopterSonde's horizontal plane, and the axis Z is a measure of the tilt of the CopterSonde and it is denoted by γ. The magnitude and direction of the projection of v onto the plane XY of I can be directly associated with the wind speed and direction, respectively. Subsequently, the wind direction with respect to the CopterSonde's true heading (X axis) is 1λ=arccos⁡(nyz⋅(eθ×eϕ)xy|nyz|⋅|(eθ×eϕ)xy|), and after some vector operations and math simplifications, λ results in 2λ=arctan⁡(-cos⁡θsin⁡ϕsin⁡θcos⁡ϕ), where eθ=cos⁡θ0-sin⁡θ and eϕ=0cos⁡ϕsin⁡ϕ, as similarly deduced by . Additionally, the CopterSonde is constantly measuring its heading with respect to the true north by means of an onboard compass which is denoted by ψ. The sum of ψ with the computed angle λ gives the absolute wind direction. However, since the algorithm commands the CopterSonde to turn into the wind, the angle λ tends to zero, and consequently, ψ tends to the absolute wind direction. It should be noted that the wind vane algorithm works only when the CopterSonde is horizontally steady, and the tilt angle γ is produced only as a result of the CopterSonde compensating for wind.

The autopilot board executes the wind vane code and calculates the wind direction λ at a frequency rate of 10 Hz. Despite the fast computation of λ, it is better to take the average over a time period of a sequence of λs to filter out any perturbations caused by the rotor wash and other undesired aerodynamic disturbance. The time-averaged angle λ‾ is then inserted directly into the autopilot flight control algorithm as a yaw command, turning the CopterSonde about its yaw axis to face it into the wind. After several bench tests of the code with the simulator, the time interval for the average was set to 5 s or 50 samples. It was found that a shorter time interval causes the CopterSonde to oscillate about its yaw axis, while a longer time interval outputs wind direction estimates at a very slow rate.

The CopterSonde was designed to be a modular system; in particular, the payload has its own detachable compartment capable of operating independently from the main body of the CopterSonde. This feature is particularly useful for calibration and maintenance purposes. The CopterSonde's shell is divided into two pieces: the front shell and the back shell, as shown in Fig. . The back shell's main purpose is to protect the vital components inside the CopterSonde such as the battery and the autopilot. Moreover, it makes the CopterSonde more aerodynamic and its rear vertical fin helps with orienting the CopterSonde into the wind. The sensor compartment is located in the front shell of the CopterSonde. The shells are exchangeable, which provides the option to have a variety of shells with different sensor configurations for a broad range of experiments. The dimensions of the compartment are approximately 100×95×70 mm and can carry a maximum payload weight of 205 g and a minimum of 90 g (not including the shell). The CopterSonde's center of gravity (CG) without the shells is slightly rear-shifted with respect to its center. Under the minimum payload weight configuration, the CG is located at the center of the CopterSonde, whereas the CG is slightly forward under maximum payload weight configuration. None of these configurations seem to compromise the CopterSonde's stability or performance to any noticeable degree. The current version of the CopterSonde's front shell is outfitted with a set of three thermistors and three humidity sensors for redundant thermodynamic sampling. Figure  shows the location of the sensors and the fan inside the L-shaped duct incorporated into the front shell. The sensors were placed in an inverted “V”-shape pattern to prevent the sensors from sampling heated and disturbed air produced by the wake and self-heating of the upstream sensors. The fan was calibrated to draw air across the sensors at a constant speed of 12 ms-1. It automatically switches on (off) after takeoff (before landing) to protect the sensors from dust close to the ground. Further analysis and considerations about the sensor placement on the CopterSonde can be found in and .

The additional meteorological sensors are the iMet-XF bead thermistor and the IST HYT-271 capacitive humidity sensor, because of their small sizes, low cost, and ease of programming. Pressure is measured by a single MS5611 sensor built-in the Pixhawk autopilot board which is also used for altitude control. The specifications of the sensors are summarized in Table . The temperature and humidity sensors were programmed to use the I2C communication protocol to exchange data streams with the Pixhawk autopilot board. Only two connecting cables are required to transfer data between the payload and the CopterSonde's autopilot board. Additionally, the shell carries its own battery that powers the sensors and the fan. It has enough energy for four consecutive flights before needing to be replaced. It should also be mentioned that the front shell can be redesigned to fit other types of sensors, such trace gas monitors, provided that the sensors can fit within the payload dimensions and weight limitations.

The design of the CopterSonde was mostly driven by the meteorological sampling needs, one of them being the acquisition of data representative of the environment. Therefore, a thorough study of the airflow around the CopterSonde and its payload was conducted. The aim of the experiment was to mitigate undesired modifications to the sampled air resulting from the CopterSonde body and characterize any possible bias. To address this challenge, a digital 3-D model of the CopterSonde body was created, and subsequently, flow simulations were conducted using CFD included in the SolidWorks^® program.

First, real observations were conducted with the CopterSonde in windy conditions at the Kessler Atmospheric and Ecological Field Station (KAEFS) in Purcell, Oklahoma, USA, located 30 km southwest of the OU Norman campus. The CopterSonde completed several hovers while oriented into and perpendicular to the wind. The flights were conducted next to the Washington Oklahoma Mesonet 10 m meteorological tower WASH;, which recorded the atmospheric conditions close to the surface. The external temperature of the CopterSonde body and shell was also recorded using an infrared thermometer. The observations were then used as initial conditions for the flow simulation. A steady state in the simulation was reached after approximately 0.8 s of simulated physical time, which took about 8 h of computing time. Figure  shows a cross-section view of the airflow's temperature with streamlines denoting the air's path. This helped in identifying the air trajectory, patterns, and sources of heat around the CopterSonde and across the sensors. Noting that the blue color is the environmental temperature, a heat aura can be discerned around the CopterSonde. Despite the hottest location being only 1 K higher than the environmental temperature, it can be seen that the region in the air flow corresponding to the smallest temperature perturbation occurs inside the L-shaped duct. Based on the simulation results, the sensing elements are exposed to the cleanest airflow relative to the CopterSonde's surroundings. However, other environmental factors, such as the solar radiation, were missing in the simulation. Solar radiation has a significant contribution to data contamination as found by . Work is underway to produce a more accurate modeling of the CopterSonde that includes impacts from environmental factors.

Every new build of the CopterSonde undergoes a series of calibration and validation steps. First, the CopterSonde's autopilot system is tuned to achieve the most stable flight under a wide range of weather conditions. Detecting any undesired oscillations or sluggishness in its behavior in this stage is crucial. If the tuning is poor, the errors can potentially propagate and impact wind estimations or produce other serious consequences in performance. Another advantage of the modularity of the CopterSonde's sensor payload is that it can be calibrated as a single system instead of its individual components or sensors. The calibration and validation of the sensor package are described in the following sections.

The CopterSonde concept extends beyond the aircraft and also includes data handling and distribution systems for the end user. There are three primary components related to the distribution of sensor data from the CopterSonde. The first component is the sensor package within the flight controller that is responsible for collecting the measurement data, logging to a local log file, and transmitting the collected data to the GCS. Sensor data are packed into a custom MAVLink message and transmitted through a telemetry radio to the GCS using ArduPilot's standard messaging platform. The second component is the GCS software, which processes the MAVLink message, plots the live data on various graphical interfaces for visual feedback, and forwards the data to the online cloud service. The final component is the data processing systems located within the cloud computing service Microsoft^® Azure. Sensor data are transmitted via the Advanced Message Queueing Protocol (AMQP) to a preprocessing Azure Service Bus, where it then waits to be dequeued by one of the data processing virtual machines. The virtual machines analyze the ingested data, store the results in a cloud database, and finally pass the data to a post-processing Service Bus for interested subscribers.

The CopterSonde system stores the collected data in multiple ways, which is convenient for creating backups and retrieving the data in any circumstance. While in flight, the CopterSonde is continuously broadcasting data through the telemetry radio to the GCS, which stores and uploads the data to the Azure cloud. The processed data are transferred directly from the post-processing Service Bus to the subscribers, as mentioned in Sect. . For example, the public web application http://wxuas.com (last access: 1 November 2019) is subscribed to such service and it was created to distribute the CopterSonde's data. The web application then forwards the information directly to connected web clients via web sockets. This enables a live view of the CopterSonde's data for any remote user via the web during flight operations. Another method of data storage is by using the included removable micro-SD card mounted on the Pixhawk board. This method is preferred over the streamed data for post-processing because it stores higher-temporal-resolution data. However, retrieving the data from the SD card requires partially taking the CopterSonde apart which hinders the goal of making the system autonomous. Therefore, there are plans to improve data transmission and increase the temporal resolution of the streamed data to reduce dependence on the SD card transfer. Some useful products generated based on data acquired by the CopterSonde are the Skew-T log-p figures and time–height temperature contours shown in Figs.  and , respectively. Post-processing and visualization of CopterSonde data are performed in Python using the open-source packages NumPy , Matplotlib , MetPy , and SciPy . These two types of visualizations allow atmospheric scientists to understand small-scale ABL processes in frameworks they are already familiar with from radiosondes and ground-based remote sensors.

The LAPSE-RATE campaign took place in the San Luis Valley in south-central Colorado. This campaign aimed to study various atmospheric phenomena, such as the morning ABL transition, convection initiation, and valley drainage flows, using spatially distributed weather sensing UAS from different research institutes . Between the three CopterSonde platforms flown during LAPSE-RATE, there were over 200 successful flights over a span of 5 consecutive days. CASS flew at two different locations each day, one of them situated at the Moffat Consolidated School where the Collaborative Lower Atmospheric Mobile Profiling System CLAMPS; was running and launching radiosondes every couple of hours. CLAMPS contains an Atmospheric Emitted Radiance Interferometer (AERI), a HATPRO microwave radiometer (MWR), and a Halo Photonics scanning Doppler wind lidar. The AERI and MWR data were combined and run through the AERI optimal estimation AERIoe; retrieval to obtain temperature and moisture profiles of the boundary layer. The Doppler lidar performed velocity azimuth display (VAD) scans and vertical stares. CLAMPS also has the ability to store helium tanks and launch radiosondes.

The CASS team operated under a certificate of authorization (COA) granted by the FAA that allowed flights to a maximum altitude of 914 m a.g.l. The CopterSonde flew vertical profile flights with an ascent rate of 3.5 ms-1 and a descent rate of 6 ms-1. It took approximately 8 min to perform a profile, allowing for a 15 min cadence when needed to observe features of interest. The cadence was determined by the meteorological conditions observed from the live data readout discussed in Sect. .

Sample data from the Moffat site are shown in Fig. a–c compared to both co-located radiosondes and remote sensors from CLAMPS. During this time period, winds were less than 5 ms-1 throughout the CopterSonde profile (Fig. a). The CopterSonde generally estimated the wind speed to be approximately 2 ms-1 less than both the radiosonde and the VAD from the CLAMPS Doppler lidar. The CopterSonde did successfully capture a directional shear layer around 750 m (Fig. b). It should be noted that the wind direction data presented in Fig. b reveal better agreement between the CopterSonde and VAD than the CopterSonde and radiosonde. Since the radiosonde drifts with the wind, only the measurements from the CopterSonde and VAD represent true vertical profiles. Moreover, at lower altitudes, below 200 m a.g.l., there is considerable scatter in the wind direction measurements from the CopterSonde and VAD, likely because of the low speeds. The radiosonde data have been smoothed over height. Here, again, there is better agreement between the CopterSonde and VAD. The swinging motion of the instrument package on a radiosonde can produce erroneous result at low altitudes after release. Though the wind speed bias falls outside the stated accuracy above (±0.6 ms-1), it is a consistent bias and can be corrected. Work is ongoing to determine a calibration procedure for calculating the coefficients in Eq. () for an ascending rotary-wing UAS, as opposed to a hovering rotary-wing UAS (see Sect. ). For this case study, the temperature measured by the CopterSonde is nearly identical to the radiosonde temperature (Fig. c).

In the months following LAPSE-RATE, a local, OU-coordinated campaign, known as Flux Capacitor, was organized to evaluate the CopterSonde over a full 24 h diurnal cycle. This campaign was held at KAEFS. Similar to LAPSE-RATE, CLAMPS was deployed and radiosondes were launched while flights with the CopterSonde were conducted. During this campaign, OU operated under a COA that allowed flights to a maximum altitude of 1500 m, though the team usually could no longer see the CopterSonde past approximately 1200 m and thus was forced to descend to comply with the FAA rule on maintaining visual line of sight on the platform. Figure d–f show example data from this campaign.

This time period was characterized by a strong nocturnal low-level jet (Fig. d) with wind speeds approaching 25 ms-1 at around 500 m. Due to safety concerns, the CopterSonde was commanded to descend if the pitch was greater than 30∘ due to the high winds. This generally occurred when the wind speed approached 22 ms-1. Again, the CopterSonde estimated wind speeds to be slightly lower (still 2 ms-1) than what was measured by the radiosonde (Fig. d). As mentioned before, this is likely due to the coefficients used for the wind estimation Eq. (). Additionally, the temperatures measured by the CopterSonde agree with the observations from the radiosonde (Fig. f).

More detailed statistical analyses of how the CopterSonde performs compared to radiosondes and remote sensors can be found in .