Application Of Calibration Transfer Techniques Between Different Mid-Infrared Spectrometers/Modules To Improve Accuracy In Estimating Soil Properties

ORCID

Gamagedara: https://orcid.org/0000-0002-6826-8420; Feng: https://orcid.org/0000-0001-7783-1644; Tagert: https://orcid.org/0000-0002-0745-2099; Martins: https://orcid.org/0000-0003-3802-0368; Wijewardane: https://orcid.org/0000-0001-8962-9451

MSU Affiliation

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

Creation Date

2026-06-30

Abstract

Diffuse reflectance spectroscopy offers a rapid and cost-effective alternative to traditional soil property measurement. Advances in spectrometer technologies have enhanced portability and affordability, expanding their use for soil property estimation. However, developing training datasets for new spectrometers is expensive and time-consuming. Leveraging existing spectral datasets is crucial, yet variations between different spectrometers reduce prediction accuracy. To address this issue, we conducted model training and testing using Mississippi and Texas datasets from the USDA National Soil Survey Center–Kellogg Soil Survey Laboratory mid-infrared (MIR) spectral library (n = 2564) and regional dataset (n = 1521) across four Fourier-transform MIR spectrometers/modules. We assessed calibration transfer techniques using preprocessing (individual/combinations) and spectral/model transfer for predicting soil properties. Among preprocessing techniques, combination of first derivative with Savitzky–Golay, baseline correction (BC), standard normal variate (SNV), and combination of BC, SNV outperformed others, though no single approach was optimal for all properties. Spectral/model transfer techniques such as external parameter orthogonalization and spiking effectively harmonized predictions, while slope-bias correction, direct standardization, and piecewise direct standardization showed limited success. A combined approach of BC and SNV spiking significantly improved model performance across spectrometers/modules and soil properties. On average across all the soil properties, the mean R2 improvement compared to models trained without calibration transfer was 0.354 when using the spectral library for training and regional dataset for testing, and 0.401 when using regional dataset for both training and testing. This study demonstrated that existing spectral datasets can be effectively used for new spectrometers with calibration transfer, allowing real-time and field-scale soil property measurement.

Publication Date

10-21-2025

Publication Title

Soil Science Society of America Journal

Publisher

Wiley

Rights

© 2025 The Author(s)

Link to Full Text

Share

COinS
 

Digital Object Identifier (DOI)

https://doi.org/10.1002/saj2.70147