Assessment of Different VisNIR and MIR Spectroscopic Techniques and the Potential of Calibration Transfer Between MIR Laboratory and Portable Instruments to Estimate Soil Properties

ORCID

Wijewardane: https://orcid.org/0000-0001-8962-9451; Cox: https://orcid.org/0000-0002-1197-3567; Zhang: https://orcid.org/0000-0001-9654-3859

MSU Affiliation

College of Agriculture and Life Sciences; Department of Plant and Soil Sciences; Department of Agricultural and Biological Engineering; James Worth Bagley College of Engineering

Creation Date

2026-06-30

Abstract

Spectroscopic analysis of soil using visible-near infrared (VisNIR) and mid infrared (MIR) regions is a rapid, low-cost, and nondestructive tool which has the potential of substituting or complementing conventional laboratory methods. Numerous studies have used different data acquisition, preprocessing, and modeling techniques to predict different soil attributes, but their impact on prediction accuracies was not consistent. In addition, instrumental disparities prohibit the application of models from laboratory to portable spectrometers limiting field applications. The goal of this study was to enable the field application of spectroscopic techniques for soil sensing using portable spectrometers. To this end, three objectives were defined: (i) to evaluate the impact of preprocessing and modeling algorithms, (ii) to compare the different spectral regions and portable versus laboratory spectrometers, and (iii) to evaluate the potential of different calibration transfer approaches to eliminate the instrumental impact in MIR region. A total of 474 soil samples were collected, air dried, ground, and sieved to obtain <  2 mm fraction followed by scanning with five spectrometers. Four preprocessing techniques (no preprocessing, Savitzky-Golay, standard normal variate, multiplicative scatter correction) were compared for prediction accuracy. Four different modeling techniques (partial least square regression (PLSR), support vector regression, random forest, and artificial neural network) were used to build and validate the models. Results revealed that PLSR outperformed all other nonlinear modeling techniques and preprocessing was not required to calibrate robust and reliable models. In general, the MIR region outperformed the VisNIR region while portable instruments performed on par with laboratory instrumentation. Four calibration transfer methods (external parameter orthogonalization (EPO), direct standardization (DS), slope bias and spiking with extra weights) were deployed to evaluate the transferability of the models between laboratory and portable spectrometers in the MIR region. Extra weighted spiking consistently yielded superior performance in correcting instrumental disparities in the spectra with EPO and DS showing significant variability in prediction accuracy across different properties.

Publication Date

3-19-2025

Publication Title

Soil and Tillage Research

Publisher

Elsevier

Rights

© 2026 The Authors

Link to Full Text

Share

COinS
 

Digital Object Identifier (DOI)

https://doi.org/10.1016/j.still.2025.106555