On the model–based estimation of backscatter texture from SAR image texture for area–extensive scenes

This study addresses the discrimination and characterization of sea-ice scenes using synthetic aperture radar (SAR) image texture. We develop a model for a second-order statistic, the autocorrelation function. This model is based on a phenomenological model of the radar scattering from sea ice in a marginal ice zone (MIZ) and a model of the SAR imaging system suitable for airborne geometries. These models are based on our understanding of the scene and system and become implicit assumptions in the subsequent texture analysis. In order to examine the validity of these assumptions we have developed an experimental methodology which is useful when doing texture analysis outside the domain of sea-ice scenes and airborne SAR geometries. This methodology involves two sets of experiments focused on the scene and system models. The first set of experiments demonstrated that the simple second-order SAR imaging model was appropriate for the airborne imaging geometry used in the collection of the SAR data. It was further shown that the sea ice responds to the SAR as diffuse targets and that the statistics of imagery with large processed bandwidths are very sensitive to errors in system focus. The second set of experiments pointed out the difficulty in testing for fully developed speckle with SAR image data. These experiments also demonstrated that several of the ice types did indeed respond to the SAR as non-Gaussian targets. A small percentage of pixels in the complex image data appeared to have correlated in-phase and quadrature components. However, the evidence allowing us to infer that partly developed speckle is the cause of these correlations is weak and ambiguous.