Abstract

The role of soil in the global carbon cycle and carbon–climate feedback mechanisms has attracted considerable interest in recent decades. Consequently, development of simple, rapid, and inexpensive methods to support the studies on carbon dynamics in soil is of interest. Near-infrared spectroscopy (NIRS) has emerged as a rapid and cost-effective method for measurements of soil properties. The aim of this study was to develop and validate a predictive model for δ13C values using NIRS in various soil profiles across Chile. Eleven sites were selected in the range of 30° to 50° S. These sites represent different soil moisture and soil temperature regimes, clay mineralogies, parent materials, and climates; in addition, they have prairie vegetation and contain C3-type vegetation. Air-dried soil samples were scanned in the NIR range at a resolution of 4 cm−1. The carbon isotopic composition, expressed as δ13C relative to the Vienna Pee Dee Belemnite standard, was analysed using an elemental analyser–isotope ratio mass spectrometer system. A prediction model for δ13C values based on NIRS data was developed through a partial least-squares regression (PLS) model using ten latent variables. A second model was generated using a random forest (RF) approach. The model performances were acceptable. The RF model provided the best results. The values of the root mean square error of prediction for the validation runs for δ13C obtained using the PLS and RF models were 1.38‰, and 1.15‰, respectively. These model performances indicate that NIRS can be used to predict δ13C for the selected dataset. The results of this study support the use of NIRS as a predictive method in soil analyses and as a nondestructive waste-free method for studies on carbon dynamics in soil.