Analysis and modeling of relationships between salinity and spectral reflectance of soils

In support of remote sensing applications for monitoring processes of the Earth system, research was conducted to analyze the basic spectral response related to moisture and salt content of soils in the visible and reflective infrared wavelengths. Surface samples of 6 Mollisol pedons were examined. Three soils were from salt-affected and three soils from salt-free areas. The moisture experiment was conducted on 213 soil samples with known moisture content (%) by weight. A subset of 118 samples was prepared to be in air dry, pF4.2, pF3.4, pF2.0 and pF1.0 moisture tensions. To study the effect of salinization and desalinization, 162 spectral measurements were taken on 9 treatments of the 6 soils with 3 replications in air dry conditions. Bidirectional reflectance factor measurements (BRF) were obtained in laboratory with an Exotech 20C spectroradiometer over the 0.515-2.325 $\mu$m spectral range in 0.01 $\mu$m increments. A test of homogeneity of the covariance matrices on the odd and even number of bands indicated a significant difference in variation of some of the 0.01 $\mu$m bands. To reduce the high spectral dimensionality of the 181 bands, a set of 9 bands for the moisture and a set of 12 bands for the salinization/desalinization study was selected based on the zero-points of averaged first few eigenvectors obtained from principal component analysis. The effect of moisture on soil reflectance was modeled by using simple and stepwise multivariate analysis. It was found that (i) BRF has a negative logarithmic relationship with moisture content (% by weight), and a positive logarithmic relationship with moisture content expressed as log(tension). (ii) The correlation coefficient (R$\sp2)$ for moisture (%) was always greater than R$\sp2$ for moisture tension for each spectral band. (iii) The best correlations, R$\sp2$ = 0.82 for moisture % and 0.77 for moisture tension, were given by 2.045-2.055 m infrared band of the non-water-absorption bands. (iv) The estimation of moisture (%) was improved (R$\sp2$ = 0.94) the most by including three spectral bands in the multivariate model. The analysis of spectral separability was conducted on the BRF of soils grouped at 5 moisture tensions and on BRF grouped by soils at each moisture tension level. Soils in pF3.4 and pF2.0 conditions showed the greatest spectral separability. The variations in reflectance showed a strong relationship with the soil variations related to salinization and desalinization. It was found that although the broad bands, such as the simulated MSS bands separated 75-79% of soil/treatment combinations, a large number of additional soil/treatment combinations were distinguished by narrow bands selected on the basis of uncorrelated spectral features. A set of 12 bands was chosen for further studies with an overall 92.9% separability of all examined combinations of the 6 soils and 9 treatments