Phase unwrapping (PU) is a key step in interferometric synthetic aperture radar (InSAR) surface monitoring, which is directly related to the accuracy of the line-of-sight (LOS) displacement observations. Underground coal mining generally results in large gradient displacements, producing dense and aliasing fringes in interferograms. In this case, it is challenging to accurately unwrap interferometric phases using traditional methods. This letter proposes an iterative algorithm to unwrap interferometric phases over underground coal mining areas. First, differential interferogram is unwrapped by the minimum cost flow (MCF) method to obtain unwrapped phases at the edge of the mining subsidence basin. Then, a 2-D elliptical Gaussian function, which could approximately describe the spatial shape of mining subsidence basin, is used to iteratively reduce the phase gradients of interferograms, so that residual wrapped phases at more pixel in subsidence basin can be unwrapped. The presented method was tested over Datong coal mine, China, based on 16 TerraSAR-X acquisitions. The result suggests that the accuracy of time-series displacements along the LOS direction solved from the unwrapped phases using the new PU method is 1.1 cm, indicating an improvement of 73.2% compared with that of using the traditional MCF method.