Date of Graduation


Document Type


Degree Name

Doctor of Philosophy in Crop, Soil & Environmental Sciences (PhD)

Degree Level



Crop, Soil & Environmental Sciences


David M. Miller

Committee Member

Andy Mauromoustakos

Second Committee Member

Trenton Roberts

Third Committee Member

Phillip R. Owens

Fourth Committee Member

Zamir Libohova


calibration optimization, chemometrics, Climate change, DRIFT MIR, mid-infrared spectroscopy, soil organic carbon, soil spectroscopy


Resource-efficient techniques for accurate soil carbon estimation are necessary to satisfy the increasing demand for spatiotemporal data. In the last thirty years, mid-infrared (MIR) soil spectroscopy has developed as an accurate, rapid, cost-effective, and non-destructive technique for soil organic carbon (SOC) analysis. In soil spectroscopy, a calibration model relates spectral data to a corresponding measured soil property and is subsequently used to predict this value from new spectral data. Various optimization techniques have been used to improve the statistical performance of calibrations; however, there is little consensus on the conditions that make these techniques effective. The objectives of this research were to (1) assess current trends in optimization techniques and conditions that render them effective for SOC (%) prediction, (2) validate the use of subsetting by environmental and soil attributes as an effective optimization technique, and (3) evaluate the effectiveness of taxonomic and mineralogic criteria and spiking as effective optimization techniques for spectral library transfer. For the first objective, a review of current optimization techniques, including the selection of calibration set size and the construction of targeted calibration models through subsetting and spiking, was performed. A decision chart for the selection of optimization techniques for spectroscopic modeling of SOC (%) was constructed and general guidance for the application of these techniques to small and large soil spectral libraries (SSLs) was provided. For the second objective, a dataset of MIR spectra and corresponding SOC (%) measurements from Nebraska and Kansas was extracted from the USDA-NRCS National Soil Survey Center-Kellogg Soil Survey Laboratory (KSSL) MIR SSL. The dataset was subset based on environmental criteria (climate, topography), soil attribute criteria (wetland, SOC (%), parent material type), and a combination of both for the construction of calibration models. Subset models reduced the prediction error by 13 to 56% relative to the full set model. Moreover, subset models constructed using 2 to 80% of the full set observations resulted in as or more accurate predictions than the full set. For the third objective, fractions of the KSSL library based on taxonomic (orders and suborders) and mineralogic (carbonate content) criteria and spiking were used to construct calibration models to predict SOC (%) in Cul de Sac, Haiti. Subsetting by suborders improved predictive performance over subsetting by orders, but neither model resulted in a desirable prediction error (≤0.40%). Spiking the general library calibration sets with 25 Cul de Sac observations produced the most desirable and reliable predictions. In addition, the spiked models outperformed the Cul de Sac model in terms of reduced prediction error. The research conducted suggests that subsetting can be an effective optimization technique and that subsetting alone or in combination with spiking are effective optimization techniques for library transfer using the KSSL MIR SSL.