Three-dimensional phase unwrapping is one of the key steps of time-series interferometric synthetic aperture radar (InSAR) technology. The unwrapping results directly affect the accuracy of time-series InSAR land subsidence monitoring. Aiming at the problem of the whole period unwrapping error caused by phase undersampling in areas with serious land subsidence and large change of terrain slope, a weighted least squares phase unwrapping algorithm based on frequency confidence is proposed, which replaces the weighted least squares phase unwrapping algorithm based on phase gradient in the time-dependent three-dimensional phase. By improving the accuracy of phase slope change estimation, the accuracy and stability of spatio-temporal three-dimensional phase unwrapping can be improved. The land subsidence results of the typical area in Beijing are obtained by using the improved spatio-temporal three-dimensional phase unwrapping method. Compared with the classical spatio-temporal three-dimensional phase unwrapping method, the improved method can obtain higher accuracy of land subsidence monitoring results. Especially for the subsidence funnel region where the slope changes greatly and the decorrelation exists obviously, the subsidence monitoring accuracy has improved significantly.