Radiometric Normalization for Multimode Image Comparison

Intercomparison of backscatter collected by SAR sensors at heterogeneous radar look angles gives rise to highly variable ground areas being associated with each pixel location within a radar geometry (slant or ground range) image. Many elements within a digital elevation model (in map geometry) can be mapped to a single location in the radar image (range / Doppler coordinates). An image simulation technique uses a faceted high resolution elevation model to integrate all backscatter returned to each range and Doppler location in the radar image (incorporating knowledge of local radar shadow). Modelling the imaging process in this manner, a map of illuminated area is produced in radar geometry, and used to normalize the true backscatter returned by the radar sensor. Although radar shadow must be considered specially, no extraordinary treatment is required of layover regions, as they are correctly accounted for by integration of the illuminated area. The image simulation approach improves on the conventional consideration of 2D incidence angles, as the 3D configuration defining the illuminated area (in both the range and azimuth dimensions) is captured. RADARSAT images acquired over Switzerland are used to demonstrate the benefit of such normalization for thematic interpretation. A high resolution digital elevation model (DEM) with 25m pixel spacing is used as input to the image simulation. The deterioration of the normalization with progressively poorer input DEMs is studied empirically to gauge the required DEM resolution for acceptable normalization of images acquired over pre-alpine topography.