An active bistatic LIDAR system operating through atmospheric turbulence is considered. Illumination field is assumed to be an electromagnetic Gaussian-Schell model beam. Target surface is modeled as a combination of isotropic phase screen governed by Gaussian statistics, to account for its roughness, and a Gaussian lens to account for its size and radius of curvature. With the help of a recently developed tensor method for propagation of stochastic electromagnetic beams through ABCD systems and random media we examine the evolution of states of coherence and polarization of the beam. In the case of unresolved flat (planar) target we show that by comparing coherence and polarization properties of the illumination beam and of the return beam it is possible to predict the typical roughness of the target surface. (C) 2010 Elsevier B.V. All rights reserved.