The aerial measurement of wind speed and direction is important for the development of the low-altitude economy, meteorology, climate research, and renewable energy systems. Existing UAV-based wind measurements, whether instrument-based or flight-dynamic-based, consistently exhibit bias and significant errors, limiting their reliability for precise wind estimation. This study introduces a machine learning (ML) approach to improve the accuracy of the wind speed and direction estimation using UAVs. The proposed method leverages data from sensors onboard UAV platforms, combined with advanced ML algorithms trained on ground-truth measurements obtained through high-resolution LiDAR systems. The experiments reveal that incorporating a 10 s smoothing window yields a root mean square error (RMSE) value of 0.39 m/s for the wind speed (horizontal) and an even lower bias (≤0.069 m/s) when using a 60 s smoothing window, representing a marked improvement over traditional techniques. These results are particularly promising at longer smoothing windows (>50 s), where the ML-based approach achieves superior accuracy compared to LiDAR measurements. The findings underscore the potential of integrating machine learning with UAV-based wind measurement systems to achieve higher precision and reliability in wind characterization.

There was an error in the original publication