Removing the Effects of Water From Visible-Near Infrared Spectra in Soil Profiles for the Estimation of Organic Carbon

The terrestrial carbon cycle is the most important constitution and plays a prominent role in global carbon cycle, and soil carbon sequestration makes an important contribution to the global climate change. The full soil profile is a highly dynamic component of the ecosystem, with pronounced depth-dependent processing of soil organic carbon (SOC) such that accumulations and losses of carbon above ground are on different temporal trajectories than changes below ground. But many studies do not consider the spatial variability of soil properties in the vertical direction, especially in the Qinghai-Tibet Plateau, mainly because of the difficulties and expense of collecting soil material in that kind of terrain and transporting it to the laboratory. Visible near-infrared reflectance spectroscopy (Vis-NIR) is an increasingly popular measurement method that is enabling the rapid, real-time and accurate proximal sensing of soil properties, including SOC. However, the soil moisture content has been shown to affect soil spectra, might mask or alter the absorption features of SOC. EPO and PDS are two effective methods to correct soil spectra effects, but we still unknow the feasibility of those two methods on fresh profile samples. In this study, we compared EPO and PDS on a set of 26 soil cores (1 m depths and 5 cm diameter) in the Sygera Mountains on the Qinghai-Tibet Plateau, China. Spectra were acquired from fresh, vertical faces 5 cmx5 cm in the area from the centers of the cores to give 386 spectra in all. We also got the spectra and SOC contents from the 386 dry samples. The statistical models were built to predict of the SOC in the samples from the spectra by Random Forest. The bootstrap was used to assess the uncertainty of the predictions by the EPO and PDS. Our results show that PDS is an effective strategy to mitigate the effects of soil water content on vis-NIR spectra for the fresh soil core samples from arable and grassland. While EPO neither shown significantly out performed those wet core samples from grassland. There were somewhat differences along with the profile on prediction accuracy of SOC between EPO and PDS. Both EPO and PDS show significantly available on surface layers of samples from arable and grassland. The EPO and PDS illustrated the dependence of land use type and soil depth. Our work would be a benefit to therapid and accurate estimation of the vertical partitioning of SOC content in the field in alpine mountains using Vis-NIR spectroscopy.