A method is derived for determining the dependence of radar backscatter on incidence angle that is applicable to the region corresponding to a particular radar image. The method is based on enforcing mathematical consistency between the frequency distribution of the image's pixel signals (histogram of DN values with suitable normalizations) and a one-dimensional frequency distribution of slope component, as might be obtained from a radar or laser altimetry profile in or near the area imaged. In order to achieve a unique solution, the auxiliary assumption is made that the two-dimensional frequency distribution of slope is isotropic. The backscatter is not derived in absolute units. The method is developed in such a way as to separate the reflectance function from the pixel-signal transfer characteristic. However, these two sources of variation are distinguishable only on the basis of a weak dependence on the azimuthal component of slope; therefore such an approach can be expected to be ill-conditioned unless the revision of the transfer characteristic is limited to the determination of an additive instrumental background level. The altimetry profile does not have to be registered in the image, and the statistical nature of the approach minimizes pixel noise effects and the effects of a disparity between the resolutions of the image and the altimetry profile, except in the wings of the distribution where low-number statistics preclude accuracy anyway. The problem of dealing with unknown slope components perpendicular to the profiling traverse, which besets the one-to-one comparison between individual slope components and pixel-signal values, disappears in the present approach. In order to test the resulting algorithm, an artificial radar image was generated from the digitized topographic map of the Lake Champlain West quadrangle in the Adirondack Mountains, U.S.A., using an arbitrarily selected reflectance function. From the same map, a one-dimensional frequency distribution of slope component was extracted. The algorithm recaptured the original reflectance function to the degree that, for the central 90% of the data, the discrepancy translates to a RMS slope error of 0.1 ???. For the central 99% of the data, the maximum error translates to 1 ???; at the absolute extremes of the data the error grows to 6 ???. ?? 1988 Kluwer Academic Publishers.
Additional publication details
Radarclinometry: Bootstrapping the radar reflectance function from the image pixel-signal frequency distribution and an altimetry profile