Relationship of Landsat spectral data with earth surface features on a semiarid rangeland

Abstract

Correlation coefficients and multiple-linear regression equations were attempted to relate earth surface features to the four Landsat MSS bands and to transformations of those bands. Twenty-three sample sites with four replications were used. Two regression analyses were attempted using all sample sites and using a twenty-sample subgroup with a 10YR Munsell hue. Correlation coefficients showed that only simple linear regression equations could be derived. Per cent silt, per cent clay, and square-root of Munsell value were the only variables that had regression equations developed in both studies. A regression equation in the total sample study was also derived for the CIE x coordinate. Spectral band transformations were less highly correlated with the field variables than the highest correlated spectral band. All spectral variables were ineffective in determining changes in vegetation density at the low vegetative cover (< 25%) found in this study area. The best single discriminator of the field variables was band four. Using a Mann-Whitney U test, Ruin Granite could be separated from Pinal Schist in all four bands and soils with 10YR Munsell hues could be separated from redder hue soils of 5YR and 7.5YR in the visible bands.