Variability of Laboratory Measured Soil Lines of Soils from Southeastern Brazil

Abstract

Linear relationships (soil lines) between the conventional red (R) and near-infrared (NIR), and between unconventional NIR reflectance bands were investigated for the effects of spectral positioning and widths of approximately simulated bands of some broad- and narrow-band sensors, and for the influence of the chemical constituents and moisture in the samples. The investigation was based on laboratory spectra of air-dried samples from soil profiles from southeastern Brazil, an experiment for the chemical removal of organic matter, and on a wetting process of the soil samples. The relationships were evaluated for a general soil line and for the individual lines of some soil types, under air-dried condition, through the determination of the line parameters, dispersion statistics, and principal components analysis (PCA). Soil lines obtained with broad or narrow NIR bands positioned at shorter wavelengths, and consequently at smaller distance differences between the pair of bands, presented better fitted lines than the ones obtained with NIR bands located at longer wavelengths and at larger distances. Compared with the broad bands, spectrally better positioned narrow NIR/R bands produced closely similar results because of the propitious effect of the larger bandwidth. However, unconventional NIR/NIR bands presented better fitted lines than the R/NIR bands. The inherent interplay between organic matter and iron oxides is accountable for the variability in the linear parameters of the individual lines. The general line for the samples considered is composed of nonparallel segments of the different soil types, with a tendency for an increased obliquity in the low albedo soils. Soil types with very low albedo present line slope values below 1, poor correlations, but small root mean square (RMS) values, whereas soil types with moderate and high albedo present line slope values above 1, strong correlations, and large RMS values. The former is rich in organic matter and iron oxides whereas the latter has more transparent components, such as quartz, or less opaque substances. However, in soil types with a lesser content of organic matter, the iron spectral features are enhanced and variations in the shape of the spectra are produced because of the larger reflectance increase in the R band than in the NIR interval. The maximum spectral contrast resulting from this compositional interplay is assessed when the NIR band is positioned at longer wavelengths. Similarly, soil wetting introduces nonlinear spectral changes that also affect the characteristics of the soil lines. The analysis of field measured reflectance spectra derived from an experiment designed to monitor the growth of a bean crop indicates that the shift of the NIR band towards shorter wavelengths produces not only better fitted soil lines but also stronger vegetation signals. The enhancement of the vegetation signal obtained with the displacement of the NIR band results from the steeper descending slope towards longer wavelengths of the NIR plateau observed in the vegetation spectra. The results place constraints on an unsupported choice of bands to derive vegetation indices, especially those of the advanced hyperspectral imaging sensors. The most recommended pair of NIR-R bands for crop vegetation indices is the one from LANDSAT-TM, closely followed by JERS-OPS and SPOT-HRV, or from narrow NIR-R bands positioned within these ranges at spectral intervals not affected by atmospheric attenuation.