Conventional models in distributed video coding (DVC) consider the correlation noise to be distributed independently from the realization of the side-information. This paper introduces a novel model, of which the standard deviation depends spatially on the realization of the side-information. The performance penalty in video coding caused by side-information-independency assumptions is theoretical quantified and experimentally confirmed. Furthermore, inspired by the spatial side-information-dependency of the proposed model, a novel approach for estimating the correlation channel from the partial knowledge of it is introduced. The proposed technique is incorporated into a spatial-domain unidirectional DVC system, providing state-of-the-art performance.