ON THE UNICITY OF POLARIMETRIC TARGET SCATTERING DECOMPOSITION R. Touzi Canada Centre for Remote Sensing Canada Centre for Mapping and Earth Observation 560 Rochester, Ottawa, Ont. K1A0E4 email: ridha.touzi@ccrs.nrcan.gc.ca ABSTRACT The objectives of target decomposition theory is to express the average scattering mechanism as the sum of independent elements and to associate a physical mechanism with each component [1, 2, 3, 6]. Two categories of incoherent target decomposition (ICTD may be distinguished; the model-based decompositions (MBD) firstly introduced by Freeman and Durden [7], and the eigenvector based decomposition introduced by Cloude and Barnes [4, 5]. MBD supposes that target observed scattering can be modeled as the linear sum of scattering that can be represented by models of the physical scattering process [7]. The Freeman decomposition [7] assumes that target scattering can be modeled as the linear sum of surface, double-bounce and volume scattering. Yamaguchi et al [8] added helix as a fourth component. Orientation angle compensation is then applied for roll invariant decomposition [9-11]. Unfortunately, MBD may not lead to a unique [20] scattering decomposition, because of the presence of a negative component brought out in [8, 9, 12]. Since the demonstration of the existence of MBD negative powers [9, 12], many “new” MBDs [12-14] have appeared, as an extension of Freeman decomposition with additional scattering models to minimize the negative component. This leads to many MBDs with a target scattering decomposition that does depend on the a-priori scattering model used, as discussed in [20]. Besides, the MBD assumption of reflection symmetry may lead to biased measure of scattering contribution when a processing window of small number of independent samples is used, as discussed in [20]. Large processing window are required to cancel the terms related to scattering reflection symmetry assumption [20]. Since all the most popular MBDs are currently applied with small processing window size (5x5, 7x7 or 9x9 on 1-look image), the results obtained with MBD should be used with big care [20]. In particular, the MBD decomposition of non stationary scattering, such as the one backscattered from urban features, might be a serious issue. Small processing window is required to apply MBD under local stationary conditions, and this leads to seriously biased MBD estimation of individual scattering contribution [20]. The eigenvector-baseddecomposition (EVBD) [4, 5] is based on the characteristic decomposition of target scattering covariance matrix. The latter breaks, in the monostatic