Nonlinear methods for the identification of 2D models with long-range dependence

In this paper we propose an anisotropic model for 2D fractional noise generated from a Gaussian mixture distribution. We propose two identification methods for this model: the first one is a forward modeling approach based on even order cumulants of the field, while the second one is an inverse modeling approach where the synthesis filter is inverted in order to obtain an estimation of the driving noise. Simulation results are presented which show the effectiveness and the robustness of the proposed methods.