Prediction of organic and inorganic carbon contents in soil: Vis-NIR vs. MIR spectroscopy

Increase of atmospheric carbon (C) content has been regarded as main reason of global climate warming. This trend can be reduced by soil C sequestration. However, spatial variance of soil properties brings about the difficulty of accurately mapping C contents by traditional site-by-site chemical measurement. It is necessary to develop instant and economical means for soil C measurement. This study used spectroscopic techniques to predict organic carbon (OC) and inorganic carbon (IC) contents in soil at a farm scale with an aim of comparing the calibration performance between a visible-near-infrared (Vis-NIR) spectrophotometer with recorded spectral range of 400-2500nm and a mid-infrared (MIR) Fourier Transform spectrophotometer with recorded spectral range of 2500-25000nm (or 4000-400cm^-1). A total of 100 soil samples were collected from an experimental farm at Silsoe, Bedfordshire, United Kingdom. The spectra were subjected to a partial least squares regression (PLSR) with leave-one-out cross validation to build calibration models for OC and IC contents. Validation results showed that the PLSR models developed for OC yielded coefficient of determination (R²) of 0.90 and 0.95, residual predictive deviation (RPD) of 3.16 and 4.43 for Vis-NIR spectra and MIR spectra, respectively. Although both instruments produced low accuracy for IC prediction, the MIR model (R²=0.63 and RPD=1.69) outperformed the Vis-NIR model (R²=0.31 and RPD=1.24). MIR spectrophotometer seemed more accurate for predicting soil C content than Vis-NIR spectrophotometer under laboratory condition. However, due to its robust and compact design without moving parts, Vis-NIR spectrophotometer is promising for portable or in-field measurement of C contents.