In this paper, a problem of adjacency matrix tuning in sensor networks is considered. The object of estimating is linear discrete time-varying system with several measured outputs. Each measurement may be dropout with given probability. The external disturbance belongs to class of zero mean disturbances with positive anisotropy level of extended vector. The goal is to modify initial adjacency matrix with respect to fixed estimation model to ensure minimum value of anisotropic norm for input-to-error system. Contrary to standard H2 and H∞ methods, anisotropy-based approach for estimation problem demonstrates more reliable results when stochastic parameters of disturbance are not well defined. The numerical example demonstrates efficiency of supported algorithm.