Developed is a new method of moment estimation applied to a wind profiler that measures Doppler spectra as a function of range in order to reject some clutter that corrupts the wind measurements. This method is based on lattice model inspired by both self-organization and Hopfield model in the neural network field and detects atmospheric signals in Doppler spectra to estimate radial velocity. Radial velocity estimated using this method from clutter-contaminated spectral data measured by a boundary layer radar showed a very smooth change along range. It was found that (1) horizontal winds converted from the radial velocity were more continuous in space and in time than those from conventional estimates and clearly showed wind shear ; and (2) horizontal wind showed good agreement with rawinsonde-wind.