Machine classification of freshwater ice types from landsat—I digital data using ice albedos as training sets

Abstract

A method has been used to convert ground-measured freshwater ice albedos to Landsat-1 digital numbers indirectly corrected for atmospheric attenuation and path radiance. The method has been tested by using the digital numbers as training sets to machine classify Landsat-1 digital ice cover data from a portion of northern Green Bay imaged on 13 February 1975. Results showed that the conversion algorithm produced digital numbers that, when used as training sets, classified approximately 50% of the test area. Differences between the classified test area and a previous machine classification of the scene, as well as the percentage of unclassified area caused by lack of data for some surface types, points to the need for a more comprehensive, well-documented library of signatures representing Great Lakes ice types.