Band selection procedure for multispectral scanners.

Advances in high spectral resolution sensors and in data handling capabilities are enabling development of greatly improved remote-sensing devices for resource monitoring, so that design trade-offs are required. A methodology for optimizing selection of spectral bands for multispectral instruments such as those on the LANDSAT series of satellites is described. The method is applied to a collection of laboratory and outdoor spectra of natural and artificial materials. These reflectance spectra represent the visible and near-infrared spectral ranges at high (0.01-μm) spectral resolution. For most natural materials 15-25 spectral bands appear to be sufficient to describe spectral variability, whereas description of minerals and some artificial substances may require double this number of bands.