Consideration of peak parameters derived from continuum-removed spectra to predict extractable nutrients in soils with visible and near-infrared diffuse reflectance spectroscopy (VNIR-DRS)

Publication date: November 2014
Source:Geoderma, Volumes 232–234
Author(s): Radim Vašát , Radka Kodešová , Luboš Borůvka , Aleš Klement , Ondřej Jakšík , Asa Gholizadeh
Visible and near-infrared diffuse reflectance spectroscopy (VNIR-DRS) provides a rapid and inexpensive tool for simultaneous prediction of a variety of different soil properties. Usually, some sophisticated multivariate mathematical or statistical methods are employed in order to extract the required information from the raw spectrum scan. For this purpose especially the partial least squares regression (PLSR) is the most frequently used algorithm. This method generally benefits from complexity, with which the soil spectra are treated. But interestingly, also techniques which focus on only one specific spectral feature, such as simple linear regression dealing with single continuum-removed spectra (CRS) value at selected wavelength, can often provide competitive results too. Such methods rely on known spectral signature of spectrally active soil components. In this study focusing on laboratory soil spectroscopy, we attempted to enhance the potential of CRS by taking into account all possible peaks as derived from CRS and relating their basic parameters, i.e. area, width and depth to soil properties employing the multiple linear regression (MLR) technique. On top of that comparison to PLSR was performed to evaluate the ability of the presented method. Nine measured soil properties on total 97 topsoil samples, were Mehlich 3 extractable elements Ca, Cu, Fe, K, Mg, Mn, P, and Zn and soil pH in CaCl2 extract. In seven cases (Ca, Cu, Fe, Mn, P, Zn and pH), of which three (Ca, Cu and Zn) were predicted reliably accurately (0.50<R² cv <0.80) and the rest four (Fe, Mn, P and pH) only poorly (R² cv <0.50), better results (the differences in R² cv up to 0.1) were obtained with the presented methodology compared against PLSR. For K and Mg, it was clear that K was predicted accurately while the prediction of Mg was not satisfactory, slightly better results (the differences in R² cv were 0.02 and 0.05, respectively) were achieved with PLSR against the presented method. We further concluded that content of clay, soil organic matter (SOM), and soil color were the main driving forces behind the prediction using soil spectroscopy in this particular case. The study indicated that MLR based on CRS peak parameters could be an alternative method in quantitative prediction of different soil properties using VNIR-DRS.