Robust radiometric terrain correction for SAR image comparisons

We demonstrate a robust technique for radiometric terrain correction, whereby terrain-induced modulations of the radiometry of SAR imagery are modelled and corrected. The resulting normalized images may be more easily compared with other data sets acquired at different incidence angles, even opposing look directions. We begin by reviewing the radar equation, pointing out simplifications often made to reduce the complexity of calculating the backscatter coefficient, normalized either by ground area (sigma 0 ), or illuminated area projected into the look direction (gamma 0 ). The integral over the illuminated area is often approximated by a scale factor modelling a simple planar slope, departing only slightly from "ideal" flat terrain: for gamma 0 , the radar brightness (beta 0 ) is normalized via modulation with the tangent of the local incidence angle. We quantify the radiometric errors introduced by ignoring terrain variations, comparing results based on (a) a robust radar image simulation-based approach properly modelling variations in local illuminated area, and (b) an ellipsoidal Earth assumption. A second simplification often made in solving for backscatter using the radar equation is the assumption that the local antenna gain does not vary significantly from a simple model draping the antenna gain pattern (AGP) across an Earth ellipsoid, returning the local antenna gain as a function of slant range alone. In reality, the AGP is draped across the Earth's rolling terrain retrieval of properly calibrated backscatter values should model these variations and compensate for them: although smaller than the errors caused by not property modelling variations in local illuminated area, they can be significant. We use well-calibrated and annotated ENVISAT ASAR images acquired over Switzerland to show how robust radiometric terrain correction, incorporating models for the variations of local illuminated area with terrain enables calibrated mixture of imagery acquired at differing incidence angles. Only robust retrieval of backscatter values enables such inter-mode comparisons - a capability that significantly reduces the required revisit time for monitoring changes to the radar backscatter. In conclusion, we describe a technique for combining a set of terrain-geocoded and radiometrically calibrated images derived from ascending and descending passes and multiple incidence angles to create composite radar backscatter maps. At each point, the contribution of each