Spatial-based feature extraction for estimating radial wind speed in Doppler radar spectra

The power density spectra of wind profiler are usually contaminated by persistent ground clutter. The radial wind speed estimated in such spectral data can produce a considerable error if each spectrum is processed independently. For the clutter-elimination purpose, we propose an algorithm for estimating radial wind speed that takes into account a spatial relationship within range-gated spectra. The candidates for the atmospheric echo in each spectrum are firstly determined by detecting all local power density maxima in the smoothed and normalized spectra. The spatial relationship is then exploited by linking those local maxima across gates, with respect to a continuity criterion. A number of features can be extracted from each candidate and a neural network is subsequently used to identify the local maximum most likely reflecting the radial atmospheric velocity.