Wind profiler Doppler power spectrum segmentation using U-Net

Abstract

Wind profiler radars can continually and effectively probe the atmosphere to obtain the Doppler power spectrum of ambient air motion. In addition to ambient air motion, the Doppler power spectrum used to get wind estimates may contain atmospheric and non-atmospheric disturbances. The wind estimations might be biased as a result of these disruptions. Accurate detection of ambient air motion, even in the presence of disturbances, is essential to reduce the impact of these biases. The Doppler power spectrum can be segmented using cutting-edge deep learning models to retrieve ambient air motion. In this work, we used one of the finest deep learning models, U-net, to segment the Doppler power spectrum. The proposed method’s performance evaluation shows promising results in segmenting and retrieving ambient air motion.