The periodogram maximum estimator is usually employed in coherent Doppler wind Lidar. However, the estimates are biased in the far-field with low signal-to-noise region, and the wind speed errors will increase. The baseline- drift of the range gate power spectra is one of the error items, which biases the benchmark of spectral peak distribution and interferences the peak- frequency detection. In order to correct the drifting, the smoothness prior approach based on the regularization penalized least squares is introduced to estimate the spectral baseline. In the atmospheric wind speed measurements, the result shows that the proposed approach can remove the drift baseline effectively, improve the wind speed estimation precision in far-fields significantly and increase the detection range of the wind Lidar ultimately.