This paper proposes a novel method for complete model-based decomposition of Pol-InSAR data. By decomposing the interferometric cross-coherence matrix, the approach separates scattering mechanisms (odd-bounce, double-bounce, and volume scattering) for each pixel. Unlike traditional methods using the polarimetric coherence matrix, this approach tackles the challenge of a non-Hermitian cross-coherence matrix, leading to non-real eigenvalues. A new approach is introduced to address this non-symmetry issue. The retrieved parameters include both magnitude and phase information, ultimately aiming to classify each pixel and determine its vertical position.