In situ measurements of organic carbon in soil profiles using vis-NIR spectroscopy on the Qinghai-Tibet plateau.

We wish to estimate the amount of carbon (C) stored in the soil at high altitudes, for which there is little information. Collecting and transporting large numbers of soil samples from such terrain are difficult, and we have therefore evaluated the feasibility of scanning with visible near-infrared (vis-NIR) spectroscopy in situ for the rapid measurement of the soil in the field. We took 28 cores (≈1 m depth and 5 cm diameter) of soil at altitudes from 2900 to 4500 m in the Sygera Mountains on the Qinghai-Tibet Plateau, China. Spectra were acquired from fresh, vertical faces 5 × 5 cm in area from the centers of the cores to give 413 spectra in all. The raw spectra were pretreated by several methods to remove noise, and statistical models were built to predict of the organic C in the samples from the spectra by partial least-squares regression (PLSR) and least-squares support vector machine (LS-SVM). The bootstrap was used to assess the uncertainty of the predictions by the several combinations of pretreatment and models. The predictions by LS-SVM from the field spectra, for which R(2) = 0.81, the root-mean-square error RMSE = 8.40, and the ratio of the interquartile distance RPIQ = 2.66, were comparable to the PLSR predictions from the laboratory spectra (R(2) = 0.85, RMSE = 7.28, RPIQ = 3.09). We conclude that vis-NIR scanning in situ in the field is a sufficiently accurate rapid means of estimating the concentration of organic C in soil profiles in this high region and perhaps elsewhere.