Diffuse reflectance spectroscopy for the characterization of calcareous glacial till soils from north central Montana

Abstract

Diffuse reflective spectroscopy (DRS) is a method of soil carbon (C) quantification. In this study, the Vis-NIR (350 - 2500 nm) and MIR (2500-25000 nm) regions were evaluated to determine respective predictive accuracies of soil organic and inorganic carbon (SOC and SIC, respectively). The dataset included 315 soil samples of glacial till origin, obtained from six independent farm sites within the Golden Triangle region of Montana, with depths ranging from 0-100 cm. For Vis-NIR analysis, Local vs. Regional vs. Global calibration sets were compared by six-fold cross validation by site of C predictions developed by Partial Least Squares (PLS) regression and Boosted Regression Trees (BRT). First derivative spectral data was used along with four preparation methods: (i) field moist and (ii) dry cores, (iii) 2-mm sieved ("Sieved") and (iv) milled samples (<200-um, "Milled") were used to evaluate the potential application to in-situ analysis. The most accurate SOC predictions were from Milled samples using a Local calibration set. SOC predictions were a result of SOM electronic absorptions within the visible region.

The most accurate SIC predictions were from Sieved samples with a Local calibration set. Within the MIR analysis, spectral transformations, KBr dilution and Regional vs. Local calibrations were evaluated using the same independent validation procedure of PLS calibrations. Spectral transformations included: absorbance (AB), first derivative (D) and Kubelka-Munk (KM) from samples prepared both with and without KBr dilution. The most accurate SOC predictions were obtained from models developed with the D transformation of Local spectral data. SOC predictions were a result of SOM, clay and carbonate absorptions. The most accurate SIC prediction was from the KM transformation of Local spectral data. KBr diluted samples gave comparable regional predictive accuracy to neat samples. Specular reflection was found in neat sample spectral signatures; with Local core addition, neat samples built more accurate prediction models despite these distortions. In the end, it was determined that the MIR region provides more accurate predictions of soil C for calcareous glacial till soils of north central Montana. But, the Vis-NIR region presents an accurate method while existing as the less expensive and more efficient route.