Differential Interferometry Synthetic Aperture Radar (D-InSAR) is a hot-spot technology for detecting large-scale ground deformation, among which, two-pass D-InSAR is regarded as a classical algorithm. Two-pass D-InSAR is an algorithm with high stability and simple process flows, whose precision can always be influenced by the accuracy of DEM, spatial decoherence, and factors like that. Therefore, this paper introduces an improved two-pass D-InSAR algorithm to solve these problems: 1) apriori filters in azimuth direction is used to recover the spatial and temporal coherence; 2) three-level coregistration strategy is used to improve the matching precision. By comparing the vertical deformation chart with traditional two-pass D-InSAR process flow without the proposed two steps, it is proved that the improved algorithm has better stability and accuracy than traditional process flow, and has a good prospect in urban land subsidence detecting field in the future. Through the experiment, the land subsidence in the center area of Shanghai has been successfully detected using the improved two-pass D-InSAR algorithm which importing two images of Envisat/ASAR data from 2003 to 2005. Compared with ground level measuring data with the same spatial and temporal attributes that proved by statistic index (t-test method), the improved algorithm in this paper can give viable and reliable results. Combined with historical results inverted by InSAR technology, the land subsidence condition of Shanghai has rebounded and is accelerating again, which should be paid enough attention.