Definition of Spectrally Separable Classes for Soil Survey Research

Spectral classes defined by a Euclidean distance criterion have shown similarity to soil classes as mapped by the National Cooperative Soil Survey (local, state, and federal governments cooperating). However, two problems have occurred in past studies: 1) The number of spectral classes defined by the researcher is arbitrary. 2) Spectral properties of nonvegetated soils are modified by rainfall, cultivation, and other factors not directly related to the classes of interest for soil survey purposes. This research outlines a procedure for defining spectral classes such that the differences between classes can be quantified. It also facilitates determination of a number of classes such that the classes are spectrally discriminable. This is accomplished by partitioning the data into many classes and then combining similar spectral classes on the basis of appropriate criteria. Multispectral data were collected over a 12-mile flightline in White County, Indiana, in connection with the 1971 Corn Blight Watch Experiment. Data were collected in May by the University of Michigan airborne scanning spectrometer at an altitude of 5000 feet. Spectral maps resulting from the analysis were compared to existing soil surveys of the National Cooperative Soil Survey. This method should help determine the extent to which spectral properties of soil surfaces can be associated with morphologic and topographic differences of interest to soil surveyors engaged in operational soil mapping