Retrieval of forest spatial pattern from SAR images

The spatial structure presented by forest stands and openings is of interest to scientists studying forest ecosystem dynamics. The ability to characterize forest spatial properties in terms of patch size and surface type is useful for initializing ecosystem models as well as, understanding the processes contributing to forest development. In this study, forest units with low biomass were classified from synthetic aperture radar images into several categories based on their spatial structure and different scattering mechanisms. In addition to non-forest classes several classes were identified with varying tree heights and/or tree densities that reflect natural and anthropogenic induced conditions. Optimal variables derived from Stokes matrices of P, L and C-band AIRSAR images were selected based on the separabilities between designated classes. After image classification, the number, area, and perimeter of each patch for these categories were calculated. The spatial patterns were also categorized in terms of simple spatial statistics including fractal dimensions