The present study's authors have tested many airborne delivery mechanisms for deploying meteorological sensors into tornado core flows over the past 2 decades. These include air cannons, parachuted sensor packs dropped by remote-controlled aircraft, and ground-deployed inflatable plastic bubbles to be ingested by the surface inflow of a tornado. After experimenting with many different delivery methods, it was theorized that to obtain meteorological data by a sensor from within a tornado core flow, the speed of model rockets would be required to penetrate subsiding air that is often proximal to a mature tornado . Descending air is often near or even surrounding a supercell tornado because these tornadoes most often occur near strong gradients of vertical velocity .

To address the needed observations, the authors designed and built a custom-engineered probe (Fig. ) and firmware for the present study. Real-time data reception and location tracking were deemed necessary to ensure that the loss of the probe, due to the hostile environment or inability to recover, did not impact data collection. The software was designed to permit multiple probes to be airborne concurrently.

This lightweight probe has onboard miniaturized sensors for pressure, temperature, humidity, GPS, and a three-axis inertial measurement unit (IMU). The rocket's pseudo-Lagrangian flight characteristic and small payload capacity necessitated a compact surface-mount design (2.5 × 7.6 cm) with a total mass of 30 g (17 g plus 13 g battery). The probe's onboard sensors were carefully selected based on criteria for wide environmental operating ranges, sampling rates, accuracy, mass, and power efficiency (Table ). A high sampling rate of 10 Hz was compatible with all sensors, excluding the GPS, which has a maximum rate of 5 Hz.

The GPS sensor selected for the probe was the ORG1411 “Nano Hornet” by OriginGPS. This device has a small form factor and mass while maintaining a GPS fix in accelerations of up to 39.2 m s-2. This sensor has an inbuilt patch antenna which provides weight savings compared to an external antenna.

Similarly, the BME280 sensor by Bosch Sensortec was determined to be an ideal barometric pressure and humidity sensor, given its small mass, size, and performance. The higher pressure sampling rate was advantageous for detecting gravity waves or other fine-scale pressure oscillations within the tornado and parent mesocyclone by comparing GPS altitude to pressure. However, it was determined that the temperature measurement could be improved with the addition of the SI7053 temperature sensor by Silicon Labs. This provided improvements in accuracy and thermodynamic response time (τ63%) over the BME280 (Table ).

An IMU was included on the probe to measure the forces and accelerations experienced inside the tornado and the probe's orientation. The nine-axis BNO085 is a miniature IMU by Hillcrest Labs that provides these data from an incorporated three-axis magnetometer, accelerometer, and gyroscope. The miniature size and mass of the BNO085 are ideal for flight within a tornado. Using the IMU sensor-fusion of the magnetometer, accelerometer, and gyroscope, the pitch, yaw, and roll angles were derived. Information about orientation and forces imparted on the probe helps identify turbulence and transitions between shear layers. After manufacturing, a software routine was developed to calibrate the IMU and store the data on the electrically erasable, programmable read-only memory (EEPROM), as stresses built up in the soldering process can offset the factory calibration of micro-electromechanical system (MEMS) sensors on the die.

The probe used long-range (LoRa) radio transmission to transmit and receive data to and from a ground station. LoRa is a spread spectrum technology that can still work even when the signal is below the noise floor. The power required is also minimal compared to other technologies, and throughput is traded against range, with longer distances requiring less throughput . Transmission occurs in the industrial, scientific, and medical band (ISM) at 915 MHz and does not require licensing. Point-to-point LoRa was chosen over the more commonly used wide-area star-of-stars topology long-range wide-area network (LoRaWAN) due to the increased throughput and ability to maintain transmission in sparsely populated areas. A star-of-stars topology is one in which gateways relay messages between devices and a centralized network server.

Time-division multiple access (TDMA) was implemented with regular framing pulses from a ground station, allowing multiple probes to be tracked simultaneously. 1 Hz data transmissions allow the complete sensor data to be transmitted concurrently over LoRa for nine airborne sensors for distances of up to 100 km. The TDMA protocol was modified such that non-reception of the framing pulse did not result in an immediate loss of synchronization and that opportunistic reception could continue in the event of a loss of synchronization.

A 1/4 wave monopole antenna was chosen for the probe and coiled to reduce size. This antenna configuration allowed for an omnidirectional radiation pattern such that the probe's orientation in flight would not impact the communication range.

The electronics were waterproofed using a silicone conformal coating to protect from unpredictable behavior due to current leakage attributable to moisture. The BME280 and SI7053 sensors were protected from this coating during assembly to avoid damaging the BME280 polymer due to volatile chemicals and a reduction of the thermal response on both sensors due to the insulation of the coating. A permeable polytetrafluoroethylene (PTFE) barrier was placed over the port on the BME280 to avoid damage from water and condensing moisture during flight as is typically applied to relative humidity polymer-based MEMS sensors. Power was provided by a small rechargeable 350 mA h-1 lithium-ion polymer (LiPo) battery.

The mobile ground station consisted of a Raspberry Pi Zero W and a custom LoRa daughterboard with an inter-integrated circuit (I2C) organic light-emitting diode (OLED) display and antenna. A 1/4 wave monopole omnidirectional antenna was used instead of an increased-range directional antenna to simplify tracking the probe in a severe weather environment.

The station contained custom software on an SD card and provided a web server over a Wi-Fi network. In addition, the station broadcasted the TDMA framing pulse, handled the communication, stored data from the sensors, and provided the user interface for monitoring the sensors in real time.

Accessing the interface was via a web browser on a mobile device or computer, and an Apple iPad was used for this purpose. The bearing and distance were calculated from the chase vehicle to the probe by comparing the onboard GPS location of the iPad to the GPS location received from the probe. A 10 A h-1 rechargeable battery pack provided power.

A custom-modified Magnum Sport Loader dual-motor model rocket from Quest Aerospace was used to launch the probe into the tornado (Fig. ) with a pseudo-Lagrangian drifter deployment (Fig. ). The dual-motor design enhanced thrust, range, and payload capacity.

High velocity was required to pass through the subsidence often situated near an occluded tornado ; therefore, higher-thrust class-D ammonium perchlorate motors (Aerotech D21-7T 18 mm) were used for each of the dual motors. The motor uses Blue Thunder propellant, has a mass of 25 g, and has a peak thrust of 32.1 N. Due to a long 7 s delay between engine burnout and the parachute ejection charge, the rocket coasts further and can cover the distance required to transit into the tornado. The upper three stabilization fins were raised higher on the rocket body to improve stability in high winds. A single hand-sewn nylon parachute with an outside diameter of 0.61 m was used to provide resistance to debris and damage from the ejection charge while lofting in vertical high-velocity winds.

Rocket flight simulator software (RockSim Version 9) was used to optimize the center of pressure and achieve maximum flight range without tumbling. Simulations also tested how the rocket would respond to substantial winds and atmospheric conditions found in the inflow zone of a tornado. The maximum payload capacity of this customized design was determined to be 50 g with a parachute descent rate of approximately 4–6 m s-1, depending on air density. The probe was housed within the nose cone compartment with expanded polystyrene foam plugs at both ends to constrain the movement of the probe within the enclosure during flight and for thermal protection of the battery, while still permitting airflow across the sensors.

A mobile rocket launcher (Fig. ) with pan-and-tilt capability was engineered and mounted to the roof of the armored chase vehicle Dominator 3 (Fig. ). The motor and gears of the launcher were derived from the automated window lift system of a junkyard vehicle. The design incorporated a substantial blast plate and two 0.91 m launch rods to launch multiple rockets. The potential exists for a return lightning strike from a rocket's ionization trail, so a custom control panel was designed inside the Dominator 3 to aim and launch the rockets from within the protective armor (Fig. ).

The low mass of the rocket, propellent, and payload, combined with the engine size selection, results in a Federal Aviation Administration (FAA) Class 1 amateur model rocket designation . Regulations and recommendations for Class 1 rockets in the USA are governed by the FAA, National Fire Protection Association (NFPA), and National Association of Rocketry (NAR). These regulations specify that a Class 1 amateur rocket must not be launched within a prohibited or restricted area unless permission is obtained, which is determined with aviation maps . Additionally, the rocket must be suborbital, not cross into a foreign nation, be uncrewed, and “not create a hazard to persons, property, or other aircraft”.

The NFPA recommendations are adopted at a state level by most states. The minimum site dimension for a class-D motor is 152 m and should be “in a cleared area, free of tall trees, power lines, buildings, and dry brush and grass”. A launch of this small size requires a spectator distance of 9 m . NAR summarizes the NFPA recommendations and specifies that the site “must be free of tall trees, power lines, buildings, and dry brush and grass. The launcher can be anywhere on this site, and the site can include roads. Site dimensions are not tied to the expected altitude of the rockets' flights” .

The risk to aviation is negligible as pilots avoid tornadoes and thunderstorms. The low payload mass of 30 g after deployment is less than a North American robin, so the impact would be similar to a small bird strike in the improbable event of a collision. After payload separation, the probe and parachute behave similarly to other lightweight debris lofted in a tornado at the same altitude. The descent phase is controlled by a parachute and does not present an impact hazard. Prior to payload separation, the total mass of the rocket, engines, parachute, and payload is 140 g. Comparatively, a typical radiosonde weighs 250 to 500 g and poses a greater risk to aviation.

FAA, NFPA, and NAR have accounted for the risk to people, property, and aviation according to the class definitions they have applied to the hobby and have defined regulations appropriate for safety. Many amateur rocket launches of considerably larger sizes occur daily in the USA, and model rocketry has an excellent safety record. Hitting a vehicle or person is unlikely, and damage at this size and mass is expected to be negligible.

In addition to fulfilling the legal requirements for rocket launches, we also model, sufficiently test, and inspect rockets before launch. The authors keep fire extinguishers available, team members include a firefighter and a fire marshal, and launch site safety is assessed.

The deployment of rocket-borne sensors into tornadoes from the inflow region of supercells is associated with a heightened safety risk, as the path of the tornado often must be traversed to attain an ideal launch position (Fig. ). The authors followed an intensive safety protocol during this field project, in addition to adhering to all model rocket launch regulations. These safety measures included the use of a custom armored vehicle (Fig. ) to provide protection from the environment and a trained spotter in the vehicle for visual observation of the tornado and assessment of its motion prior to deployment. In order to maximize safety during field operations, the authors did not deploy from the direct path of the tornado and debris field and remained inside the vehicle when near the storm to minimize lightning danger. The team extensively rehearsed the deployment procedure prior to this field project, realizing that any inefficiencies or mistakes can have serious consequences near the path of a tornado.

The period between 17 May and 30 May 2019 was a prolific period for severe weather across the United States, with 456 tornadoes and three tornado outbreaks . A total of 22 were rated EF3, and two were rated EF4, including the Lawrence/Linwood tornado .

On the morning of 28 May, a robust upper-level trough with 500 hPa temperatures below -20 ∘C covered a large portion of the Mountain West region of North America. A strong southwesterly mid- and upper-level jet stream extended downstream over the Southern Great Plains. The southerly low-level jet stream (LLJ) of greater than 20 m s-1 (40 kn) maintained deep low-level moisture up to a stationary front across northern Kansas and Missouri. An outflow boundary (OFB) from a nocturnal mesoscale convective system (MCS) was draped from northwest to southeast across northeastern Kansas through peak heating (Fig. ) .

Throughout the day a dryline would advance east to central Kansas and Oklahoma as an additional focus for supercell development by late afternoon. Beneath the strengthening LLJ, the backed surface winds associated with the OFB in a narrow zone between Kansas City and Topeka would enhance 0–1 km storm-relative helicity to greater than 230 m2 s-2; this environment was conducive to violent tornadoes during late afternoon and evening (Fig. ). NOAA's Storm Prediction Center upgraded portions of northeastern Kansas and northern Missouri to a moderate risk for convective storms in the 13:00 UTC day 1 convective outlook with the potential for strong to violent tornadoes . An enhanced risk area for convective storms was active to the west and southwest of the moderate risk area for a dryline supercell mode but with lower tornado probabilities given weaker low-level wind shear. A total of 25 tornadoes occurred in the United States on 28 May 2019, 10 of which were in Kansas.

The parent supercell of the Lawrence/Linwood tornado initiated in the open warm sector over the elevated terrain of the Flint Hills at around 20:00 UTC and organized on approach to the OFB located just southwest of Kansas City. The first mesocyclone occlusion yielded a brief EF2 tornado at 22:55 UTC, followed immediately by the rapid development of a more significant, stronger tornado that began in southwestern Douglas County at 23:05 UTC.

The tornado quickly grew to over 1 km wide, producing EF3 damage within 10 min of tornadogenesis along United States Highway 59 just south of Lawrence. The National Weather Service damage survey confirmed a 51.2 km long track through Douglas and Leavenworth Counties. Peak EF4 damage occurred in Linwood, with maximum winds estimated at 76 m s-1 and a damage path width of over 1.6 km (Fig. ). The tornado dissipated near the town of Bonner Springs, just west of the Kansas City International Airport.

The supercell was of the high-precipitation (HP) class. Previous experience has demonstrated that a narrow zone of inflow into the northeast quadrant of a tornado is often free of destructive hail which would damage the rocket (Fig. ). The ideal launch location is also just ahead of the wind shift of the RFD gust front to ensure relatively calmer launch conditions . The easterly and northeasterly inflow winds at this location were more likely to transport the pseudo-Lagrangian drifter into a tornado.

Dominator 3 was positioned at the location of the marked star (38.898911∘ N, 95.260437∘ W, and 258 m a.s.l.) in Fig. , and the rocket launcher was aimed into the inflow region. At 23:17:33 UTC, the authors launched the rocket, and the payload was deployed. After launch, the tornado passed south of the launch location, with structural debris visible within the condensation funnel.

At 7.9 s after launch, 1 Hz real-time data revealed that the probe parachute deployed at 437 m a.g.l. and entered the northwest side of the tornado. The crew then continued north on Highway 59 in Dominator 3 to maintain communication with the probe as it ascended in the tornado and mesocyclone. At 23:25:09 UTC, communication was lost as the probe approached the tropopause (horizontal distance from the launch of 10.4 km and 10 680 m a.s.l.).

The probe continued the flight and was recovered the next day on the grounds of a church in Leavenworth, KS (39.19∘ N, 94.94∘ W), having traveled 51.5 km northeast of the launch location. Subsequently, the authors downloaded the 10 Hz data from the onboard EEPROM for the entire flight.

The probe recorded 5 Hz 3D GPS data for time, horizontal position, altitude, speed, and heading. The GPS velocity data recorded by the probe resolve the in situ wind speeds during the 30.2 min flight inside the tornado and parent mesocyclone with the assumption that the pseudo-Lagrangian drifter follows the flow with minimal probe-induced motion following the parachute deployment phase. The GPS trajectory of the probe's flight superimposed on the NWS damage track provides valuable data that could be used for insight into the 3D structure of the violent tornado and its parent mesocyclone (Fig. ). It depicts the rotation of the probe inside the tornado and then entrainment into the tilted updraft of the mesocyclone. Ascent time was rapid with a duration of 10 min from launch to the apex of the flight.

Owing to the availability of altitude data from the GPS and pressure data from the onboard sensor, it was possible to derive a launch-referenced and altitude-corrected pressure perturbation metric to measure the pressure deficit between the ambient pressure and pressures inside the tornado vortex and mesocyclone (Fig. e). The GPS data recorded maxima for the 3D speed of 85.1 m s-1 (Fig. b) and vertical velocity (w) of 65.0 m s-1 (Fig. c) at 858 and 2172 m a.s.l., respectively. Additionally, the maximum horizontal speed component of 83.9 m s-1 at 856 m a.s.l. (Fig. c) was consistent with the 3D speed maxima, indicating that it consisted of mainly a horizontal component. A maximum of 11 914 m a.s.l. was also recorded (Fig. a). The GPS heading showed rotation within the tornado (Fig. d).

A storm motion of 21 m s-1 from 220∘ was derived from radar and subtracted from the GPS position referenced to the launch location to produce an X/Y position relative to the storm. This was then used to plot storm relative GPS 3D velocity and pressure perturbation (Fig. ). The probe completed 1.5 revolutions around the tornado in nearly circular movement with a diameter of 1.6 km (Fig. a), which is in agreement with the estimated maximum damage width of 1.6 km . The 85.1 m s-1 wind speed measured by the probe is in accordance with the 76 m s-1 wind speed estimated by the damage survey , and the pressure perturbation (Fig. b) indicates a deficit in the range of 80 to 113.5 hPa, which concurs with surface pressure deficits measured in other tornadoes . Additionally, the wind speed is greater in the southeast portion of the tornado and lower in the northwest, which is attributed to the northeasterly storm motion (Fig. a).

The 39.2 m s-2 acceleration limit of the ORG1411 was surpassed for a short period during the powered rocket flight phase, which led to a loss of fix in two-dimensional position for 5 s and accuracy in altitude for 60 s. However, this still allowed valuable data to be recorded in the near-ground layer of the tornado, and the data and figures include this phase of the flight.

The operating range of the BME280 for barometric pressure extends down to 300 hPa per the Bosch datasheet; however, this flight reached a pressure altitude of 208 hPa. Subsequent cross-analysis between the GPS data and the Topeka 00Z meteorological balloon sounding confirmed the validity of barometric pressure readings below 300 hPa – due to the closeness of the Topeka sounding combined with minimal and consistent pressure perturbation at the top of the storm and for the descent. It should be noted that during the launch of the rocket prior to pseudo-Lagrangian flight the pressure was impacted by Bernoulli's principle due to the high velocity of the rocket flight.

During the probe flight, the ground station was handheld external to the side window of Dominator 3, and the vehicle resumed highway speeds. Due to debris falling outside the tornado circulation, safety concerns necessitated bringing the ground station back inside the vehicle for periods.

The ground station lost communication with the probe at a range of 14.5 km; however, later recovery of the probe and stored data demonstrated that the probe was still transmitting throughout the entire flight. The authors determined that the shorter-than-expected radio frequency range was due to the proximity of the 16-gauge ballistic steel shell of Dominator 3 and was caused by obstruction of the signal and the Faraday cage behavior.

Holes were punched into the nose cone compartment of the rocket to promote airflow across the temperature and relative humidity sensors during flight, as pseudo-Lagrangian drifting within the tornado limits the airflow relative to the probe. Additionally, with minimal probe-relative airflow, the effect of the electrical self-heating from the radio transceiver needed to be modeled. Other considerations were that the temperature within Dominator 3 before launch was higher than the external temperature and the probe's thermal mass was moved from inside the vehicle to outside for launch.

Due to operational and time constraints prior to launch, a thermal analysis of the enclosure was completed after recovery. Treatments for thermal response exist for radiosondes that exhibit similar concerns where slower ascent rates result in a lack of airflow past the sensors contained within .

Controlled experiments (Fig. ) were conducted to compute the thermal time constant (τ) of the probe within the enclosure, the probe outside the enclosure, and the self-heating effect. A fan, handheld anemometer, deep freezer, and ice-bath calibrated thermocouple were used for this purpose. Data from these controlled experiments, including additional documentation as to the processes used in testing, can be found in .

These experiments identified constants for thermal response (τ) and self-heating (S) to correct the temperature profile as measured by the probe. Additionally, the exact temperature was known at two points in time: at launch (due to the calibrated weather station on the roof of Dominator 3) and at a point in the descent where the air temperature external to the enclosure of the probe was equal to the measured temperature within the enclosure of the probe, and therefore no heat transfer was occurring to or from the sensor internal to the probe (Fig. ).

Equation () was derived to correct the thermal data based on Newton's law of cooling . Although the descent rate of the payload was known when moving as a pseudo-Lagrangian drifter, the unknown airflow around the sensor (v) could be determined to lie within a range of possible values. The sensor airflow (v) must be less than the descent rate due to the enclosure having venting but not being completely open to the environment. A meteorological balloon sounding (Topeka 00Z) was used as a reference for the environmental temperature profile by mapping the interpolated altitude in the sounding to the GPS altitude recorded by the sensor. It was found that there was a minimal range of plausible values for (v) that did not cause the corrected temperature to overshoot (v=0.07 m s-1) the Topeka profile or undershoot (v=0.27 m s-1). A midpoint value for (v) was chosen of 0.17 m s-1 (Fig. ), which provided the best curve fit between the corrected temperature and Topeka profiles. 1Tenv=Tt-S-T0-Se-Δt/τ1-e-Δt/τ, where τ is the time constant (Eq. ), Tt is the sensor temperature at time t (∘C), Tenv is the temperature of the environment (∘C), T0 is the sensor temperature at time 0 (∘C), Δt is the delta time tt-t0 (seconds), and S is the self-heating constant (4.18 ∘C). 2τ=τ0ρυ-n, where τ0 is the time constant at zero air velocity (seconds), ρ is the air density (Eq. ), v is the air velocity in relation to the sensor (m s-1), and n is the constant ranging from 0.4 (laminar air) to 0.8 (turbulent air); 0.8 is used. 3ρ=pRT, where p is the sensor pressure (pascals), R is the 287.058 (gas constant for dry air), and T is the 00Z Topeka sounding temperature (K).

Due to the complexities and cost (> USD 15 000) of the experimental setup to categorize the response for relative humidity in the enclosure, the authors decided to forego response time correction for these data; consequently, this necessitated using temperature instead of virtual temperature for the ideal gas equation (Eq. ). This impact was deemed negligible for this study as condensing moisture was not encountered by the sensor. Additionally, this data component could be categorized later if required and budget permitted. The effect of direct solar radiation on the thermal correction was considered; however, as most of the flight was within cloud and the sensor was shielded from direct solar radiation by the enclosure, it was concluded that any impacts from this would be negligible.