Uncertainty Estimation for Small-UAS Wind Measurements in Calm and Dynamic Environments

Small uncrewed aircraft systems (sUAS) are valuable instrument platforms for atmospheric research by enabling targeted in-situ data collection with high temporal and spatial resolutions. Field campaigns where fixed-wing sUAS are deployed often focus on measuring atmospheric state variables including wind, temperature, pressure, and relative humidity. The process of computing wind estimates using a multi-hole probe (MHP) requires additional sensor measurements, typically from an inertial navigation system (INS), as well as pressure, temperature, and humidity (PTH) sensors. Uncertainties associated with measurements from each of these devices contribute to the overall confidence in wind estimates. To compute the total uncertainty in the wind measurement requires propagating the combined uncertainties from all contributing sensors.

A two-step, pre-flight and post-flight, uncertainty estimation procedure is presented. The pre-flight step employs wind tunnel testing to quantify uncertainty during the MHP calibration. The post-flight step involves processing raw data and propagating uncertainty from contributing sensors to provide a metric for wind-estimation quality.

The RAAVEN sUAS has a battery-powered airframe with a 2.3-m wingspan. The base configuration carries a Black Swift Technologies MHP, two Vaisala RSS-421 PTH sensors, and an optical video camera. In this configuration, the RAAVEN has a reliable 2.5-h endurance. RAAVENs have been flown with additional sensors simultaneously including a turbulence sensor, particulate sensor, and two radiometers. To highlight the dependence of wind uncertainty quantification on the operational environment, wind data from two comparatively different deployments of the RAAVEN sUAS are analyzed. The Tracking Aerosol Convection Interactions Experiment (TRACER) campaign was conducted June-September 2022, south of Houston. A primary goal of TRACER is to understand the role of the sea breeze in governing convective development and intensity, the influence of different aerosol properties and concentrations on cloud and precipitation processes, and the development of a convective boundary layer in the morning hours and its relationship to the clouds that eventually form. Although the RAAVEN occasionally encountered light precipitation from local convection, most data were collected between 50-610 m AGL in relatively “calm” atmospheric conditions. The Targeted Observation by Radars and UAS of Supercells (TORUS) field campaign was conducted throughout the Great Plains May-June 2019 and again in May-June 2023 for the TORUS – Left-flank Intensive Experiment (TORUS-LItE) field campaign. The goal of the TORUS campaigns is to improve the conceptual model of supercell thunderstorms (the parent storms of the most destructive tornadoes) by exposing how small-scale structures within these storms might lead to tornado formation. For these campaigns, RAAVEN teams operated in the left-flank and right-flank of supercells, and in the near-inflow environment upwind of the advancing storms.

Post-flight analysis reveals trends connecting wind uncertainty to atmospheric and sUAS control variables. By examining the relatively stable and calm TRACER flight data, we find patterns directly relating wind component uncertainty to airspeed signal strength. This pattern is not observed in the dynamic and gusty TORUS/TORUS-LItE data where wind component uncertainties are larger overall. This outcome is attributed to rapidly varying airspeed and high-rate maneuvers required to fly in the extreme conditions of supercells, often requiring flights through precipitation. These case studies show that measurement-dependent uncertainty analysis provides a platform for studying the relationship between measurement quality, environmental conditions, and UAS decision making. In a meteorological context, uncertainty propagation provides confidence bounds for wind measurements, providing a tool to set quantitative thresholds for filtering wind data. In an engineering context, real-time wind uncertainties can be delivered to on-board autopilot algorithms, with the aim of improving autopilot decision-making and performance in extreme weather.