Reconstruction of Three-Dimensional Wind Field from Sparse UAV Measurements

A recently developed hexarotor platform integrates a 20 Hz full‑scale sonic anemometer and dual‑resolution digital in‑line holography for simultaneous wind–particle sensing. The system measures three‑component velocity to within 5 % of a mast reference after flight‑specific fusion and propeller‑wake correction. Here, we assess the platform's capability for three-dimensional wind‑field reconstruction when only sparse UAV measurements are available. A neutral atmospheric boundary-layer large-eddy simulation (Re≈4.8×10⁷) serves as the benchmark flow. Synthetic flights are created by sweeping the UAV's sensor stencil through the LES domain while varying data sparsity, trajectory topology, and flight speed. Flow fields are reconstructed with the multi‑scale physics‑informed neural network. The solver infers velocity and pressure on a 128³ grid (≈1.6 M parameters) and self‑estimates an effective Reynolds number. Performance is quantified by normalized velocity RMSE, recovered integral length scales and spectral and structure-function correlations. Complementary field tests—conducted over a lidar-equipped profiling site further demonstrate the platform's practical viability. The study supplies actionable guidelines for designing UAV sampling campaigns and introduces a scalable framework for airborne, data-driven diagnostics relevant to wind-energy siting, aerosol transport, and cloud-microphysics research.