Abstract:

Uncrewed Aircraft Systems (UAS) complement radiosondes for boundary-layer profiling, but most UAS soundings sample at a fixed rate, under-resolving the thin inversions, hydrolapses, and wind-shift transitions that matter most for severe-weather forecasting. This project, part of the TADSS (Teaching Any Drone to Sample Smarter) framework, tested whether a UAS could anticipate a boundary-layer point-of-interest’s location from its sounding shape so far, matching it against a library of real historical radiosonde profiles to sample more densely where it matters. The approach uses K-Medoids(PAM) clustering of seasonally stratified IGRA2 radiosonde archives with a two-of-three POI voting scheme, built on the BLISS CopterSonde platform and LAAIRS ground control station at NOAA’s National Severe Storms Laboratory (NSSL), Norman, OK. A temperature-convention fix cut sounding-matching MAE from~18°C to 0.4–1.6°C; accuracy gains plateaued beyond ~15–20 medoids, showing clustering resolution — not library size — as the main constraint. POI detection validated against field data, with transition zones consistently at800–1,100 m AGL, and end-to-end matching now runs inside LAAIRS