Dependency Measures for Assessing the Covariation of Spectrally Active and Inactive Soil Properties in Diffuse Reflectance Spectroscopy

ment technique for several spectrally active soil properties (chromophores) such as sand, clay, organic C, and Fe contents. The approach is also used for estimating many spectrally inactive constituents (non-chromophores) based on the assumption of covariation between non-chromophores and chromophores. The linkage between covariation and the ability of DRS to estimate a non-chromophore has not been reported in the literature. In this study, we evaluated the covariation assumption using three dependency measures (Pearson correlation coefficient, r; biweight midcorrelation, bicor; and mutual information based adjacency, AMI), five chromophores (organic C, Fe, clay, and sand contents, and geometric mean diameter), and seven non-chromophores (pH, electrical conductivity, P, K, B, Zn, and Al contents) measured in 247 Alfisol and 249 Vertisol samples. An average dependency index (ADI) was developed for each of the three measures (ADI(r), ADI(bicor), and ADI(AMI)). The first derivative of the reflectance in conjunction with partial least squares regression was used for data modeling. Model accuracy was evaluated using residual prediction deviation (RPD). The relationships between RPD values of non-chromophores and the ADI values were examined for different chromophore groups (physical, chemical, and combined). The performance of ADI(AMI) was found to be superior to ADI(r) and ADI(bicor). The ADI(AMI) computed using chemical chromophores gave strong linear relationships (R-2 = 0.93) between ADI(AMI) and the RPD of chemical non-chromophores, suggesting that the AMI may be used as a robust dependency measure to assess the covariation of non-chromophores with chromophores in DRS.