The project to develop a line-integral approach to 2-dimensional radarclinometry and to bring it to the status of producing topographic maps from real radar images has been concluded. The final developments of the theory itself have involved a trial-and-error resolution of the curvature decision process at each integration step over range as follows: (1) Locally Indeterminate Azimuth-Azimuth Curvature is invoked if the range-directed path of integration is within 1 ??? in angle of the tangent to a local characteristic curve of the partial differential equation of radarclinometry (equivalent to a lapse in the necessity for an auxiliary curvature assumption); (2) Local Cylindricity is invoked if the local image isophote has a radius-of-curvature greater than 50 pixels; (3) Least-Squared Local Sphericity is invoked if the characteristic curve trends at greater than 70 ??? to the range direction (the auxiliary curvature assumption is becoming a sufficiently strong influence as to warrant the overconstraint), and (4) the default hypothesis, which is invoked most often, is the localization through the Euler/Lagrange equation from the calculus of variations of the global principle of minimization of the surface area of the terrain. The development of the set of line integrals into a 2-dimensional topographic surface is not practically achieved by branching the line integral at the range threshold, because the radarclinometry equations are too frequently coupled but weakly to the slope component in the direction of radar-azimuth, and under circumstances for which the powerfully influential auxiliary curvature assumption is too unrealistic. In other words, a line integration in radar-azimuth is far more frequently directed orthogonally to the local characteristic curve than is one carried out over range. Such orthogonality results in stepping the strike under the exclusive control of the curvature assumption. Instead, a quasi-surface-integration step is taken by modeling the dependence on initial strike of the gravitational potential energy of the vertical slab of terrain under the range-profile. The adopted starting strike for the range integral is the one which minimizes the gravitational potential energy. This radarclinometric method, in combination with my recently published method for determining an effective radar back-scattering function from one-dimensional slope statistics and image pixel-signal statistics, was applied to three images. First, to separate theoretical difficulties from experimental impediments, an artificial radar image was generated from a topographic map of the Lake Champlain West quadrangle in the Adirondack Mountains. Except for the regional trend in elevation, to which radarclinometry is insensitive by design, the agreement between the original and derived topography appears good. The morphologies agree and the range of relief is the same to within 4%. As an example of data of the highest quality available from space-borne radar at the present time, a SIR-B image of very rugged terrain in the coastal mountains of Oregon was similarly processed. The result, after filtering to redistribute photoclinometric errors about the two-dimensional spatial spectrum, agrees with ground truth almost as well. As an example of the worst possible data, in terms of signal-to-noise ratio and radar incidence angle (no detraction from the praise due the first high resolution space-borne radar-imaging of Venus intended), a Venera-15 image segment in Sedna Planitia just north-east of Sapho was processed, using Venera altimetry and Pioneer roughness data for slope statistics, in spite of the resolution mis-match. Considerably more trial-and-error filtering was required. The result appears plausible, but an error check is, of course, impossible. ?? 1990 Kluwer Academic Publishers.
Additional publication details
Radarclinometry of the Earth and Venus from space-shuttle and Venera-15 imagery