Multispectral land-use/land-cover model portability in multi-temporal multi-angle very high resolution imagery

摘要

Multi-temporal multi-angle data provides multiple images, collected over both time and satellite view-angle, of a single target area. In the presented research, these data are used to explore the fundamental physical distortions present in very-high spatial resolution optical data as applied to land-use/land-cover classification. This is done by creating a land-use/land-cover model in one image of the multi-temporal multi-angle data and directly applying it to the remaining images. This direct measure of model portability provides unique insight into the dependence of very-high spatial resolution land-use/land-cover classification on the atmosphere, solar illumination, and model training location.