Statistical models for polarimetric SAR data

By introducing the idea that the number of scatterers in a resolution cell fluctuates, a statistical model known as the K distribution arises, which requires two parameters and includes the exponential distribution as a special case. The arguments leading to the one-dimensional K distribution can be extended to N dimensions. This N-dimensional model for polarimetric data can be represented by a product model, in which the observed data vector can be treated as if it were the product of two independent processes. The observed responses from vegetated targets at C, L and P band are examined to assess whether they are adequately described by a Gaussian or K distribution. Before embarking on the full analysis, theoretical marginal distributions are compared with observations for single regions of three generic vegetation types, viz cereal, forest and root crop, exemplified by grass, conifer and sugarbeet. This is to illustrate that a simple Gaussian model is not adequate to describe the data, and the need to extend the analysis to the product model