A First-Guess Feature-Based Algorithm for Estimating Wind Speed in Clear-Air Doppler Radar Spectra

Abstract

Algorithms for deriving winds from profiler range-gated spectra currently rely on consensus averaging to remove outliers from the subhourly velocity estimates. For persistent ground clutter in the echo return that is stronger than the atmospheric component, consensus averaging of the spectral peak power densities fails because the peak power density is derived from the ground clutter and not the atmosphere. To negate the deleterious effects of persistent ground clutter, as well as to attempt to improve performance during periods of poor signal-to-noise ratio, an algorithm was developed that uses the local maxima in power density in each spectrum to build multiple profiles of possible radial velocity estimates from the first to last range tale. To build profiles of radial velocity estimates from a set of spectra, the spectra are smoothed, the local power density maxima are identified, chains are formed across range gates by connecting those local power density maxima that satisfy a continuity constraint, and finally profiles are built from a combination of chains by maximizing an energy function based on continuity, gate separation, and summed power density. Features based on power density and power density after half-plane subtraction are then constructed for each profile and a backpropagation neural network is subsequently used to classify the profile most likely reflecting the atmospheric state. It was found that use of this technique significantly reduced ground clutter contamination in the horizontal beam velocity estimates and improved performance at low signal-to-noise ratios for all velocity estimates.