Algebraic Synthesis of Forest Scenarios From Multibaseline PolInSAR Data

In this paper, a new methodology is proposed for the analysis of forested areas basing on multipolarimetric multibaseline synthetic aperture radar (SAR) surveys. Such a methodology is based on three hypotheses: 1) statistical uncorrelation of the different scattering mechanisms (SMs), such as ground, volume, and ground-trunk scattering; 2) independence of volumetric and temporal coherence losses of each SM on the choice of the polarimetric channel; and 3) invariance (up to a scale factor) of the average polarimetric signature of each SM with respect to the choice of the track. Under these hypotheses, the data covariance matrix can be expressed as a Sum of Kronecker Products, after which it follows that K SMs are uniquely identified by K (K - 1) real numbers. This result provides the basis to perform SM separation by employing not only model-based approaches, generally retained in literature but also model-free and hybrid approaches, while yielding the best Least Square solution given the hypothesis of K SMs. It will be shown that this approach to SM separation is consistent with the inversion procedures usually exploited in single-baseline polarimetric SAR interferometry. Experimental validation of this methodology is provided on the basis of the P-band data set relative to the forest site of Remningstorp, Sweden, acquired by German Aerospace Center's E-SAR airborne system in the framework of the European Space Agency campaign BioSAR.