Account for terrain and viewing angle

A satellite camera turns a three-dimensional scene into a two-dimensional image. Hills, roofs and valleys do not all lie on one flat surface. When the camera looks obliquely, elevated features can appear displaced relative to the positions they would have on a flat map. Some ground may be hidden behind terrain or structures.

ASAL describes both mono and stereo imaging possibilities for the Alsat-2 pair. Stereo uses observations from different viewing directions to help infer height through parallax: the relative displacement of corresponding features. It needs recognizable matches and known viewing geometry. Two files are not automatically a useful stereo pair merely because two spacecraft produced them.

Orthorectification combines image geometry with a terrain model to reduce displacement caused by relief and perspective. The British Antarctic Survey’s guidance emphasizes knowing whether a product is corrected and what elevation information was used. Terrain-model quality matters, especially in steep terrain or strongly oblique views.

Consider a ridge observed from opposite sides. A feature visible in one image may be hidden in the other. A terrain correction can relocate observed information; it cannot reveal the surface of a valley that the camera never saw. Interpolation may fill output pixels, but those filled values do not become independent observations of the hidden ground.

Terrain changes viewing geometry

The direction and height of the Sun affect shadows. Images of an unchanged building can show different dark shapes at different local times or seasons. A dark region might be shadow, water, burned vegetation or another surface; its appearance alone is not a complete classification. Geometry, spectral information and context help distinguish the possibilities.

For a fictional construction survey, compare matching ground locations and inspect acquisition times, viewing angles, terrain correction and shadow direction before declaring a new wall. If the relevant edge is hidden in one image, mark that comparison as uncertain or seek another observation. Better inference sometimes means withholding a precise boundary rather than treating every visible contrast as change.