In this paper, we develop a novel approach to nonlinearly retrieving wind field from radial wind data, which are obtained by Single Doppler radar. In general, it is impossible to uniquely recover the 3-D wind field from single Doppler radar, since the problem is under-determined. In order to overcome the difficulty, according to the characteristic of the wind field distribution, we add the piecewise stationary assumption of the wind field function for the wind retrieval. Furthermore, when the wavelets are applied to approximate the wind field, we found the additional assumption yields the sparse coefficients. This connects the introduction of the entropy function, because their minimization results sparse solution. The corresponding nonlinear optimization can be solved by a modified affine scaling methodology, which is at a lower computational cost due to the behavior of the affine scaling technique. The real experiments demonstrate that the algorithms performs well.