Atmospheric Boundary Layer Turbulence Measurements Using sUAS with Neural Network Application

View Video Presentation: https://doi.org/10.2514/6.2022-4112.vid Atmospheric boundary layer (ABL) turbulence measurements were conducted using an ultrasonic anemometer attached to a small unmanned aircraft system (sUAS) with vertical takeoff and landing capabilities. A recurrent neural network algorithm (Long Short-Term Memory - LSTM) was trained using the sUAS rotational and translational velocities as well as the raw wind measurements of the attached anemometer to predict the reference wind speed from a meteorological tower-based wind speed measurement at high frequency (5 Hz). The performance of the LSTM model was compared with the predictions of the method by [9] that implements a correction procedure (referred to as Corrected method). It was found that the wind speed predictions by the LSTM model and Corrected method were comparable, whereas the ABL turbulence predictions of the LSTM model were superior compared to the Corrected method. Predictions of the Corrected method predominantly overestimated the observed wind speed and ABL turbulence characteristics, whereas predictions of the LSTM model mainly underestimated those observations.