1. Introduction
Dramatic, high-impact weather events, such as severe thunderstorms with hail and wind, tornadoes, excessive rainfall and flooding, tropical storms, ice storms, heavy snowstorms, and blizzards have an impact of billions of dollars per year on the economy of the United States [
1,
2]. Additionally, several industries in the U.S. are quite weather sensitive, including agriculture, transportation and electric power generation and management. Definitive links are being made between climate change and the impact it is having on the occurrence of dramatic weather events [
3]. To mitigate deleterious impacts on society and its infrastructure, it is imperative that we develop innovative means of monitoring and modeling the Earth’s atmosphere. To achieve this end, we require better observations of the lower atmosphere and an effective means of incorporating these measurements into numerical weather predictions (NWP). That is, the availability of quality atmospheric observations is critical to our ability to monitor meteorological conditions and accurately forecast the weather.
Regarding atmospheric observations, a long-desired component to U.S. operational observing systems is the ability to measure vertical profiles of wind, temperature, and moisture in the lower troposphere at high spatial and temporal resolution. These so-called sounding or profiling data can be used to assess regions of thermal stratification and the degree of atmospheric static and dynamic stability, which play a role in convection initiation and maintenance of storms entering an area. This need is reflected in several recent studies, some of which provide explicit recommendations to collect more observations within the atmospheric boundary layer (ABL) in general, with a focus on vertical sampling (profiling) in particular [
4,
5,
6,
7,
8,
9]. These reports emphatically state that our currently available observing systems are not capable of providing adequately detailed profiles of temperature, moisture, and winds within the ABL.
Overall, processes in the ABL can vary dramatically over a single diurnal cycle, as depicted in
Figure 1. Although this conceptual model of the ABL is idealized, it helps to illuminate several common features of the ABL structure: mixed layer (ML), capping inversion (CI), the stable boundary layer (SBL), entrainment zone (EZ), the residual layer (RL), and so forth. Above the ABL is the free atmosphere (FA). The temperature profile corresponding to five particular “snap shots” of this idealized ABL diurnal cycle are depicted and labeled as A–E. Time (A) corresponds to nocturnal conditions characterized by a stable boundary layer near the surface with a well mixed residual layer above. Shortly after sunrise, the ABL begins to transition as depicted at times (B) and (C). The mixed layer is forming near the surface and as it continues to grow, the stable boundary layer is lifted and compressed until it later forms the entrainment zone. During mid day, the ABL is largely characterized as a mixed layer as shown at time (D). At sunset, surface cooling begins to occur, which in turn sets up the development of the stable boundary layer again, as shown at time (E). To fully characterize the ABL, measurements of the state parameters in each of these regions are needed, preferably with adequate temporal resolution to fully capture evolution of the height of the ABL and structures within the ABL.
The ABL generally provides the moisture, instability, low-level wind shear, and forcing necessary for the formation of severe storms with attendant tornadoes, hail, lightning, and hazardous winds. Forecasters regularly look at moisture advection and moisture gradient patterns when considering the potential for convection initiation [
10,
11].Within the ABL resides the storm-generated outflows that can regulate the strength and longevity of severe storms or even trigger new storms. Knowledge of these conditions is the key to improving predictions of severe weather events. The problem is that ABL properties are highly variable on mesoscale time and space scales, which are virtually undetected by current operational observing systems.
Vertical shear in the near-ground layer can also play a significant role in severe weather formation and is an important criterion when distinguishing tornadic from nontornadic supercells. Traditional observational data, such as those provided by wind profiling radars do not offer adequate vertical resolution to resolve the shear (e.g., [
12]). The most noticeable difference between nontornadic and tornadic cases is in the lower-tropospheric wind profile, specifically, the orientation of the 0–500 m shear vector with respect to the storm-relative inflow. This implies that the tornadic cases have much more streamwise horizontal vorticity in the lowest 500 m AGL [
13]. Traditional standards used for radiosonde data processing do not capture features in the lowest 500 m well, partly on account of the pendulum effect [
14]. Therefore, there is a need to sample the ABL near the surface with finer vertical resolution.
To address the need for more measurements in the ABL, advance several key recommendations listed in the latest NASA Strategic Plan [
15], and fulfill the mandates put before NOAA in the 2017 Weather Research and Forecasting Innovation Act, we must challenge ourselves to develop observing and modeling systems that transcend conventional methods. One solution is to use remote sensing technologies to sample the thermodynamic and kinematic properties of the lower atmosphere. Using funds from a COST action initiative, which allows European Union researchers to form interdisciplinary teams to address pressing scientific and societal challenges, investigations are underway to assess the potential impact of ground-based profiling on weather forecasting [
16]. This initiative focuses on the combined use of ceilometers (referred to in the paper as ALC or automatic low-power backscatter lidars/ceilometers), Doppler wind lidars, and microwave radiometers to retrieve measurements of temperature, humidity, aerosols, and wind. It has been demonstrated by Manninen et al. [
17] that Doppler wind lidars can be used to identify sources of turbulent mixing. There have been many ABL field studies in the last decades, which have relied on multiple ground-based remote sensing technology to observe the lower atmosphere. Some recent examples include those discussed in Lothon et al. [
18], Klein et al. [
19], Fernando et al. [
20] to study the ABL. Despite encouraging results, this approach has several notable disadvantages, namely, the need for multiple remote sensing systems and the cost of purchasing the equipment; power requirements; lack of data availability within fog, clouds, and precipitation; and the need to rely on indirect approaches to retrieve the desired atmospheric parameters. Moreover, wind estimates from scanning Doppler systems such as Doppler wind lidars, radar wind profilers, and sodars are all potentially affected by complex terrain (e.g., [
20,
21]).
Another emerging technology which could have a dramatic impact on observations of the lower atmosphere is unmanned aircraft systems (UAS). It has been suggested that such profiling could be achieved by small UAS assuming that autonomous flights extend at least through the depth of the boundary layer [
22]. As outlined below, instrumented UAS are expected to provide inexpensive, accurate, and controlled observations of the lower atmosphere. The advent of weather-observing UAS (WxUAS) would complement other observing systems, such as rawinsondes, towers, satellite-based remote sensors, and active and passive ground-based remote sensors. Here, we focus on how WxUAS data would complement measurements from networks of surface-based, in situ observing stations, known as mesonets.
In recent decades, numerous states have established mesonet networks to aid decision making across various sectors ranging from emergency management to agriculture to weather forecasting to transportation [
23,
24]. In general, mesonets aim to provide multi-purpose, high-quality, real-time observations. Additionally, they provide tailored outreach (typically to the K-12, university, public safety, and agriculture communities) to expand the utility of the observations. Across the U.S., there are currently 27 statewide mesonets [
23]. These networks range in size from less than 10 stations to over 175 stations, but each has a goal of providing high resolution observations to support mesoscale weather and climate monitoring. Typically, mesonets include sensors mounted on or below a 10 m tower to sample air temperature, relative humidity, winds, solar radiation, precipitation, pressure, soil temperature, and soil moisture. The New York Mesonet additionally has profiling capability at 17 sites [
25]. This is achieved using scanning Doppler lidars (3D wind) and microwave radiometers (temperature and humidity). We should also mention that the West Texas Mesonet has profiling capability at a single site through the operation of a radar wind profiler and a sodar [
26].
In
Figure 2, we present an example of the surface mesoscale wind and humidity fields on the afternoon of 26 March 2018 as recorded by the Oklahoma Mesonet [
27,
28]. A combination of varying insolation across the state and the action of a cold front, dryline, and upper-level jet resulted in significant county-by-county variability. The implications of such features on wildfire risk (including likelihood for initiation, behavior, and smoke dispersion), convective initiation, and moisture and momentum fluxes are immense, yet inadequately understood without additional measurements of the ABL.
Mesonets can provide valuable information on the spatiotemporal structure and development of events such as the one depicted in
Figure 2; however, measurements are mostly limited to the surface layer. The vertical structure remains under sampled. Therefore, forecasters rely on statistically-based parameterization schemes [
29] and basic conceptual models to envision processes acting in the vertical dimension. Providing researchers and forecasters with data that allow them to monitor the changing 3D wind, temperature, and moisture patterns would yield considerable benefits. For example, subtle changes in the strength of the capping inversion can have a profound impact on the probability of convective initiation and the chance that storms may become severe. Further, these data can be used to initialize mesoscale and thunderstorm-scale NWP models. Knowing if, when, and where the cap might “break” is paramount to anticipating where convection could be initiated. Current parameterization schemes struggle to provide adequate information on the strength of the capping inversion, mostly due to lack of observational data.
Here, we explore the prospects of extending the mesonet concept beyond the surface by including the capacity to sample the vertical structure of the Earth’s atmosphere through the use of WxUAS. That is, 2D surface observations from tower based sensors would be complemented by profiling measurements within the ABL and lower free troposphere from instrumented unmanned aerial vehicles (UAVs) launched from a network of ground stations capable of supporting these operations. As a point of clarification, we will use the terms UAV or WxUAV when referring to actual aircraft and UAS or WxUAS in the context of the complete system, including the aircraft, ground station, and other components associated with operations. The WxUAS would add a third spatial dimension to the sampling strategy of conventional mesonets and, as such, the proposed concept is referred to here as a 3D Mesonet. This framework would allow us to fill data gaps in the ABL that conventional instrumentation cannot easily or feasibly provide and facilitate the detection of complex mesoscale features embedded in weather systems. As we have noted, the New York Mesonet has implemented ground-based remote sensing platforms to obtain profiling data at select sites. However, we see WxUAS as being a less expensive alternative to this approach, albeit with several logistical challenges that need to be addressed.
In the following sections, we begin with a general overview of the 3D Mesonet concept and its potential impact on weather forecasting. Then, we discuss the development, calibration, and validation of a WxUAV designed for the needs of the 3D Mesonet along with other components needed for operations and risk mitigation. Next, we present the data visualization system and some preliminary results that highlight the unique capabilities the WxUAS has at profiling the ABL. Finally, we will outline the anticipated trajectory of the system along with future work that would be needed to transform the 3D Mesonet concept from a fundamental research question to an operational system.
3. Platform and Sensor Development
The general concept of operations for the 3D Mesonet places certain design criteria on the WxUAV to be used. Flights should be conducted autonomously or semi-autonomously with minimal human interactions. Moreover, it should consist of a vertical take-off and landing (VTOL) aircraft for the sake of docking and charging. This can be achieved using a rotary-wing aircraft or a hybrid vehicle capable of operating in a rotary-wing mode for take off and landing and then transitioning into a fixed-wing aircraft as its primary mode for data collection. Here, we present developments of a rotary-wing VTOL vehicle, known as the CopterSonde. It should be noted that the WxUAV described below is still undergoing modifications. Therefore, it should be taken as representative of the aircraft to be used as part of the 3D Mesonet. However, we are also exploring hybrid vehicles, which are under development within CASS and will be presented in forthcoming publications.
The CopterSonde series rotary-wing WxUAV was developed in-house by a CASS team of engineers and meteorologists to be both robust and optimized for atmospheric sampling. True to its name, the platforms are able to collect vertical profiles of the atmosphere like a traditional radiosonde. In addition, these platforms provide highly resolved data (<10 m) in the lower atmosphere that are not easily achievable using conventional methods. The CopterSonde (
Figure 6) is based on the HQuad500 manufactured by Lynxmotion. It is comprised of carbon fiber plates, aluminum brackets, and carbon fiber tube legs. The rotary-wing craft is electrically powered by a lithium polymer battery pack with a capacity of 6750 mAh, which provides a maximum flight endurance of 25–30 min and can allow the platform to climb to an altitude exceeding 2000 m AGL. It has been operated in Colorado to heights of 3000 m ASL. The propulsion system of the CopterSonde is comprised of four identical brushless motors outfitted with T-style carbon fiber propellers and a 35-A 4-in-1 electronic speed controller. The electronic components are protected from the elements with a custom 3D printed two-piece plastic shell, which was designed to enhance the vehicle’s aerodynamics and allow for easy access to the interior components. Additional information can be found in Greene et al. [
39].
All CopterSonde platforms utilize open-source technologies to enable autonomous operation of the vehicle. The first of these is a Pixhawk 2.1 flight controller (ProfiCNC), which is comprised of a rugged and vibrationally isolated microcontroller, a single central processing unit that executes the flight control system and also allows for communication with a large variety of sensors over a wide range of communication protocols (I2C, UART, RS232, etc.) When paired with the Here+ GPS module (ProfiCNC), the Pixhawk is able to utilize Real Time Kinematic (RTK) GPS, which can reduce the uncertainty of the location of the UAS to less than 1 m in both the horizontal and vertical if properly set-up and calibrated. The microcontroller runs the ArduPilot autopilot software suite, which is composed of navigation software on-board the vehicle. The autopilot comes with a ground station software which is used to transmit commands to the UAV and monitor its status over telemetry link. Both software packages are open-source for any developer, which allows to create highly customizable environments/configurations and the addition of code that enables for other sensor integration and special features, such as to dictate the sampling speed of the sensors, flight path of the vehicle, ascent rate, and much more.
The design of the CopterSonde has been driven by meteorological sampling needs coupled with the overall concept of operations and the logistical requirements of unattended operations. Meteorological sampling needs include both the types of measurements to be acquired and the accuracy of those measurements. For the 3D Mesonet concept, we are primarily interested in characterizing the thermodynamic and kinematic state of the lower atmosphere, namely, atmospheric pressure, temperature, and humidity along with the wind speed and direction with
sufficient temporal and vertical resolution. For reference, we present in
Table 1 the variables to be observed using the CopterSonde and their respective target accuracies. To achieve these measurement goals, it has been necessary to carefully explore how to provide adequate aspiration of the sensors, minimize the effects from direct exposure to the sun, and insure that the measurements are not being impacted by affects from the vehicle itself, such as heating. More information on sensor placement on the CopterSonde can be found in Greene et al. [
38] and Greene et al. [
39]. An intercomparison of thermodynamic and kinematic measurements from various WxUAVs, including the CopterSonde, can be found in Barbieri et al. [
40].
The CopterSonde is equipped with three Innovative Sensor Technology HYT 271 humidity sensors, three InterMet Bead Thermistors, and a MS5611 Barometer to measure pressure. The temperature and humidity sensors are located on the front of the vehicle in the 3D printed scoop designed to protect the sensors from solar radiation and serve as the front half of the platform’s protective shell [
39]. A ducted fan is placed in the bottom of the scoop and used to aspirate the sensors by pulling air over them at a constant rate. The barometer is located inside of the Pixhawk flight controller. Wind velocity and direction is calculated using an algorithm in the autopilot software, Ardupilot, which uses the pitch, yaw, and roll angle of the platform. Additionally, the rotary-wing craft uses the results of the wind algorithm to keep the vehicle facing into the wind as to allow for sampling of air undisturbed by the propellers.
Wind speed and direction is estimated as follows. The autopilot system has been configured such that the CopterSonde can orient into the direction of the incoming wind. Since the pitch, roll and yaw of the vehicle are measured, wind direction is found from the yaw angle [
39]. The wind speed is then calculated from the pitch angle using a method similar to that described by Neumann and Bartholmai [
43] and Palomaki et al. [
44]. The requisite coefficients for the wind speed calculations are obtained by calibrating the wind retrieval algorithm as the CopterSonde hovers near a meteorological tower.
6. Conclusions and Future Directions
As we have noted, there is a clear need to improve our understanding of the Earth’s ABL, a view that appears frequently in several national studies [
4,
5,
6,
7,
8]. However, due to the complexity and temporal variability of structures in the atmospheric boundary layer, most conventional atmospheric sensing methodologies, such as radiosondes, weather radar, ground-based remote sensing instruments and satellites, fall short of providing the spatial and temporal resolution needed to understand boundary layer processes.
The continuing deployment of small WxUAS to collect in situ measurements of the atmospheric state in conjunction with surface conditions will significantly expand weather observation capabilities. Having the ability to deploy networks of WxUAS could significantly enhance the safety of individuals and support commerce through improved observations and short-term forecasts of the weather and other environmental variables in the lower atmosphere. In particular, observations from a 3D Mesonet could play a critical role in helping to address several pressing scientific questions, including several outlined in the most recent National Academies’ Decadal Survey “Thriving on Our Changing Planet: A Decadal Strategy for Earth Observation from Space” [
6]. Examining the list of “most important” questions identified in the report, we find several, which could be better addressed through the availability of improved observations of the ABL, including: How do processes in the ABL relate to air–surface (land, ocean, and sea ice) coupling? What drives the timing and location of convective storms and heavy precipitation? What processes are most relevant in determining the spatiotemporal distribution of air pollutants? How is the water cycle changing and what impact does that have on droughts and pluvials? How are the energy and water cycles being affected by anthropogenic effects? Although taken from a report with a focus on Earth observations from space, they are motivated by the need to improve measurements in the lower atmosphere, regardless of methodology.
Clearly a novel approach for ABL monitoring such as WxUAS must be developed and tested to fill this data gap. Moreover, we recognize that WxUAS technology is not a panacea. Rather, it is simply a means to an end, with the end being a better understanding of of the Earth’s atmosphere and its processes. Therefore, a holistic approach should be envisioned that involves complementary observational data from a variety of sources. Additionally, carefully constructed modeling approaches should be adopted to best utilize the observational data. This has been our guiding philosophy when developing the 3D Mesonet concept.
As outlined in this discussion, we have been able to conceptualize, develop, and successfully test many of the individual components needed to deploy a 3D Mesonet station; however, there is still much to be done before completing a fully autonomous prototype system. For example, we must identify a suitable charging station for the WxUAV based on several design considerations. The WxUAV must be able to execute a precision landing on the charging station. There must be some mechanism for aligning the aircraft with an inductive or direct connection charging port or to physically swap the battery of the WxUAV. The charging station and the WxUAV should be protected from the elements when not in use. There are commercially available rotary-wing UAS, which offer autonomous charging capability; however, we must find or build one suitable to our particular aircraft and sensor suite.
Another critical consideration when designing a 3D Mesonet prototype is that of system integration. For example, the GeoFence Radar must have software in place to reliably detect and track manned aircraft. Furthermore, when it has been determined that the aircraft has entered into geofenced air space, appropriate deconfliction actions must be taken, which could involve a variety of actions depending on conditions: delay launch, return to launch area, create separation between the manned aircraft and the WxUAV, and so forth. Similar actions could be taken if a manned aircraft is detected through its ADS-B out signal. Creating system software to allow for safe and reliable operations will require considerable development and testing.
With these considerations in mind, we continue to work towards developing a prototype of a 3D Mesonet WxUAS station, which can operate unattended. In addition to the hardware, software, sensor, and system engineering challenges, we realize that there are regulatory hurdles to overcome before a 3D Mesonet can be implemented. As previously mentioned, several risk mitigation factors are already developed or in development. Additionally, we must consider the ability of the system to operate in a range of environmental conditions, including: extreme temperatures; strong winds and turbulence; and icing. The current limitations of the system are being explored. Hardening the WxUAS against adverse weather conditions is a topic of ongoing research. During field campaigns, the CopterSonde itself has been operated in a temperature range of −25° to 30°, in winds up to 25 m s, to altitudes of 10,000 ft (3050 m) MSL, and in clouds.
As WxUAS technology continues to mature and our capacity to make robust and accurate observations of the lower atmosphere grows, we must correspondingly match this development within the realm of atmospheric modeling. Once data are available from WxUAS deployments, these observations can be assimilated into NWP models along with all other available weather data to determine the extent of improvement to the model forecasts and the longevity of the impact with a focus on high impact weather events depending on the season and location. These types of modeling studies are known as Observing Simulation Experiments (OSEs) [
48,
49]. The University of Oklahoma’s Center for Analysis of Storms (CAPS) has experience using a number of NWP models, including ARPS, WRF and the designated next-generation global prediction system (FV3), including performing mesoscale OSEs [
50]. Data assimilation can be accomplished for in situ data such as the WxUAS with efficient methods such as 3DVAR analyses assimilated with Incremental Analysis Updating [
51], recently updated at CAPS to include variable-dependent timing [
52] as well as more complex ensemble-based methods such as Square-Root Ensemble Kalman Filter (EnKF).
At this stage, the 3D Mesonet concept is only a dream, but from dreams, reality can happen. Such a network would carry us closer to filling the atmospheric data gap that exists in the ABL and help us to better understand complex surface–atmosphere interactions and energy exchanges. Admittedly, as we move forward, we must also be open to other complementary and evolving sampling technologies to collect these data. Currently, however, WxUAS offer much potential. The needs of society are placing ever increasing demands on forecasters to improve the granularity and fidelity of weather predictions. This in turn requires networks of atmospheric observing networks along with the ability to integrate the observations efficiently into NWP models. There is work to be done to get there. We feel that realizing the dream of a 3D Mesonet can play a critical role in reaching these goals.