Abstract: GB-InSAR (ground-based interferometric synthetic aperture radar) plays an important role in high precision deformation measurement. The deformation measurement reliability is affected significantly by phase unwrapping accuracy. In order to solve the cumulative phase jump error caused by random jump-noise, a 3-D spatial-temporal phase unwrapping method was proposed. Firstly, the phase jump pixels were arranged to be identified based on the unwrapped phase obtained by spatial minimum cost flow, combining the difference interference phase model to construct a correction network . And then, based on the established spatial constraint equation between the jump pixel and its neighborhood pixels, together with the temporal constraint equation between two successively interferograms to correct the 2-D unwrapped phase. Finally, simulation and measurement tests were carried out. The results of simulation and data show that the proposed method can effectively correct unwrapping errors distributed in sporadic pixels, further more correct regional phase unwrapping errors, improving the deformation inversion reliability.