In this study, an anomalous atmospheric phase screen (APS) compensation in the frame of a ground-based differential synthetic aperture radar interferometry (GB-DInSAR) is investigated. The two improved methods are proposed and verified with GB-SAR data acquired at the mountain cliff exposed by a landslide. The first method adopts the piecewise multiple regression approach by assuming the inhomogeneity of the refractivity index distribution along the range direction. As for the second method, we apply nth order of two-dimensional (2D) polynomial with respect to slant range and topography height which enables the flexible fitting for the anomalous APS. In the second method, the cross-validation is employed to avoid both underfitting and overfitting problem. The applicability of both methods is validated with GB-DInSAR data which presents the anomalous APS. As a result, both methods yield significant reduction of the residual error compared to the conventional method.