Predicting Tree Species Origin of Soil Organic Carbon with Near-Infrared Reflectance Spectroscopy

Document Type


Journal/Book Title/Conference

Soil Science Society of American Journal

Publication Date







Near-infrared reflectance spectroscopy (NIRS) and partial least squares regression were used to develop prediction models for identifying the species of origin of soil organic C (SOC) in semiarid montane forests of quaking aspen (Populus tremuloides Michx.) and mixed conifers in Utah. Artificial mixtures of mineral soils (0–15 cm) sampled under pure aspen and pure conifer cover (n = 415) at four locations were divided into a calibration–validation set (n = 265) for model development and an independent validation set (n = 150) to test model robustness. Models in the 10,000 to 4000 cm−1 spectral region were developed separately with original soil spectra (OS) and organic matter spectra (OM) using the full and truncated (10th–90th percentile) sample sets. The OS models performed better than OM models, and the best OS models showed good prediction ability at the validation step, with R2 = 76%, ratio of standard deviation of reference value to standard error of prediction (RPD) = 2.1 for aspen SOC, and R2 = 74%, RPD = 2.0 for conifer SOC. Model performance decreased at independent validation (R2 = 33– 49%, RPD = 1.2–1.6), probably due to unaccounted variability of site-specific factors in SOC chemical composition within and among aspen and conifer soils. Current models are still somewhat limited for accurately predicting contributions of aspen vs. conifers in independent samples. More detailed site information, such as texture, mineralogy, geology, and land use history is needed to improve models so that they can be used to provide insight into SOC properties changes along a continuum of aspen to conifer forests in the western United States.

First Page


Last Page