A Hybrid Model for Interferometric and Polarimetric P-band Sar Imaging of Forests

In this paper, a forward model for extended covariance matrix prediction for boreal and hemi-boreal forest in Pband SAR is presented. The main product is the extended covariance matrix scaled to sigma nought on the diagonal. The input parameters consist of basic radar setup, topography, forest biome, biomass, and some model parameters. Backscatter intensities for HH, VV, and HV channels are predicted from biomass using regression based on BioSAR 2007 campaign data. The phase of the correlation between the HH and VV channels is found to be proportional to biomass and is also modelled by a regression based on BioSAR 2007 data. The coherence of HH and VV channels is found to be unrelated to biomass and is chosen to be modelled as a stochastic variable. The correlation of any co-polarized channel with HV is set to 0. The interferometric correlation values for the three channels are modelled using volume over ground (VoG) model, which is a combination of random volume over ground (RVoG), oriented volume over ground (OVoG), and elevated random volume over ground (ERVoG) models. The forward model is also evaluated against SAR data from the BioSAR 2007 campaign. Three intensity images and one complex polarimetric correlation image are created for Remningstorp (site of BioSAR 2007) from existing biomass map, DEM, and flight path information. These images are compared with the images acquired with ESAR during the BioSAR 2007 campaign and the similarities and differences are discussed. The presented forward model is able to predict backscatter with an RMSE of 1.35 dB (HV), 1.77 dB (VV), and 1.92 dB (HH). Polarimetric correlation can be predicted with magnitude and phase RMSE equal to 0.08 and 16.3 deg, respectively. A qualitative evaluation of the interferometric part is also done and it is concluded that a good setup of model parameters is neccessary to get satisfactory results.