Comparison of Metrics for the Classification of Soils Under Variable Geometrical Conditions Using Hyperspectral Data

Abstract

The objective of this letter is to find a distance metric between reflectance spectra that is not sensitive to the variations on the soil reflectance induced by the geometry of solar-view angles. This is motivated by the fact that differences between spectra measured for the same soil under different observation and illumination configurations can lead to misclassifications. Using 26 soils of different compositions simulated with Hapke's model and 92 soils of different compositions measured under 28 solar-view angle geometries in laboratory conditions, we tested three metrics, namely, root-mean-square error, spectral angle mapper, and R 2 (the coefficient of determination), and we compared their efficiency. The best results are obtained with the coefficient of determination with 93% of good classifications.