In land seismic survey, the seismic data are mostly contaminated by ground-roll, which reduces the signal-to-noise ratio and impacts the quality of the seismic data seriously. The least-squares filter is a new means to suppress ground-roll, which is different from normal domain filter. The main principle of the method is the utilization of the convolution of scanning signal and filtering factor to predict ground-roll. It is necessary to construct the objective function to calculate the filter factor. The objective function is the least-squares error of seismic data and ground-roll. In that way, the targeting functionally obtained minimum can calculate filter factors from seismic data. The last step is to continue the second ground-roll suppression by making the extracted ground-roll as the input, which can get the seismic data and wipe out the ground-roll. In this paper, nonlinear sweep signal was used to extract ground-roll, and the sweep signal's beginning-ending frequency was analyzed. The results prove that the proposed method can effectively suppress the surface wave through synthesis of field data.