Quantification of soil organic carbon using mid- and near- DRIFT spectroscopy

by Kang, Misun

New, rapid techniques to quantify the different pools of soil organic matter (SOM) are needed to improve our understanding of the dynamics and spatio-temporal variability of SOM in terrestrial ecosystems. In this study, total organic carbon (TOC) and oxidizable organic carbon (OCWB) fraction were calibrated and predicted by mid- and near-DRIFT spectroscopy in combination with partial least squares (PLS) regression method. PLS regression is a multivariate calibration method that can decompose spectral data (X) and soil property data (Y) into a new smaller set of latent variables and their scores that best describe all the variance in the data. Oxidizable organic carbon content was measured by a modified Walkley-Black method, and total organic carbon was measured by the carbon analyzer.

The floodplain and Blackland Prairie soils in Texas were used for prediction of TOC and OCWB using mid- and near-DRIFT spectroscopy. Floodplain soil is mainly composed of quartz and kaolinite, whereas Blackland Prairie soils contain high concentrations of smectitic clays and low to high concentrations of carbonate minerals. The total organic carbon of 68 soil samples from two Texas sites varied between 0.19 and 4.36 wt.% C, and the oxidizable organic carbon of 26 samples from floodplain soils was in the range of 0.05 to 1.33 wt.% C.

TOC and OCWB of soil were successfully calibrated and predicted by the PLS regression method using mid- and near-DRIFT spectroscopy. The correlation using mid-IR spectra for TOC (r = 0.96, RMSEV = 0.32 for calibration; r = 0.93, RMSEP = 0.44 for prediction) was about the same as the near-IR result (r = 0.95, RMSEV = 0.37; r = 0.93, RMSEP = 0.42). Therefore, we can also use mid-infrared region for quantification of total organic carbon in soils. The PLS1 regression model (r = 0.92) for prediction of OCWB using mid-IR spectra was more accurate than the PLS2 regression model (r = 0.90). PLS models showed better correlation with spectral data than the univariate least square regression method(r = 0.83) with TOC measured by the carbon analyzer.

This study shows that the partial least squares (PLS1) method using mid-and near-IR spectra of neat soil samples can be used to predict both total organic carbon and oxidizable carbon fraction as a fast and routine quantitative method.