PolInSAR coherence-based decomposition modeling for scattering characterization: A case study in Uttarakhand, India

Abstract

Polarimetric decomposition models such as Freeman-Durden’s three-component and Yamaguchi’s four-component models were used to extract the scattering mechanisms. In a few cases, both the models have shown overestimation of the volume scattering from the highly-dense building instead of even-bounce scattering. Deorientation process helped to reduce the overestimation of the volume scattering but didn’t accomplish completely. This paper addressed this issue by the insertion of PolInSAR coherence in the existing decomposition models. The coherence varies with features based on its temporal and volume decorrelation, hence forest results in low coherence compared to permanent scatterers helps to characterize the man-made and natural features. This paper experimented with a proposed model on the RADARSAT-2 dataset of Uttarakhand as a case study. Volume scattering was modified and extracted based on different coherence parameter assumptions. Finally, the first optimal band with spatial and temporal baseline coefficients as 1 and 0.6 thresholds gave a more reliable outcome. For qualitative analysis, the scattering mechanisms of features from different models were compared. In the proposed model, closely spaced buildings exhibited 54% and 26% dominance in the even and odd-bounce scattering while volume scattering reduced to 18%. Whereas Freeman-Durden and Yamaguchi’s model has shown dominance in volume nearly 98% and 50% while even and odd-bounce scattering were less than 1% in Freeman-Durden model and 39% and 11% in Yamaguchi’s model. Forest shows dominance in the volume scattering higher than 60% in the Freeman-Durden and the proposed models whereas 35% in the Yamaguchi’s model. Therefore, the proposed methodology successfully fused polarimetry and interferometry to overcome the ambiguity in scattering.