Abstract: Atmospheric phase (AP) is one of the main error sources in the measurement of GB-InSAR (Ground-based Interferometric Synthetic Aperture Radar). Nonlinear AP in the GB-InSAR interferogram cannot be effectively compensated with conventional parametrical models. This paper proposes an AP compensation method based on meteorological data. Firstly, based on the empirical model of refractive index, AP curves of all the PSs (Permanent Scatterers) are estimated through meteorological data. Then, the cross-correlations between the estimated AP curves and the measured interferometric phase curves are calculated. Those stable PSs are selected by setting a proper coherence threshold. Based on the stable PSs, three different interpolation fitting methods, including locally weighted scatterplot smoothing (Lowess), linear interpolation and neural network, are utilized to estimate the AP curves of all the PSs. The most appropriate interpolation fitting method is finally determined by comparing compensation results. Experimental results prove that when the PS number is enough, three methods can all effectively compensate the nonlinear AP components, or the Lowess method could achieve the best compensation performance.