Abstract
Mid-infrared (MIR) spectroscopy has received widespread interest as a method to complement traditional soil analysis. Recently available portable MIR spectrometers additionally offer potential for on-site applications, given sufficient spectral data quality. We therefore tested the performance of the Agilent 4300 Handheld FTIR (DRIFT spectra) in comparison to a Bruker Tensor 27 bench-top instrument in terms of (i) spectral quality and measurement noise quantified by wavelet analysis; (ii) accuracy of partial least squares (PLS) calibrations for soil organic carbon (SOC), total nitrogen (N), pH, clay and sand content with a repeated cross-validation analysis; and (iii) key spectral regions for these soil properties identified with a Monte Carlo spectral variable selection approach. Measurements and multivariate calibrations with the handheld device were as good as or slightly better than Bruker equipped with a DRIFT accessory, but not as accurate as with directional hemispherical reflectance (DHR) data collected with an integrating sphere. Variations in noise did not markedly affect the accuracy of multivariate PLS calibrations. Identified key spectral regions for PLS calibrations provided a good match between Agilent and Bruker DHR data, especially for SOC and N. Our findings suggest that portable FTIR instruments are a viable alternative for MIR measurements in the laboratory and offer great potential for on-site applications.
Highlights
In recent years, soil spectroscopy has been established as an efficient method to complement conventional soil analysis [1,2,3]
We focused on five soil properties that are relevant for agricultural practices and constitute key indicators of soil health and fertility: soil organic carbon (SOC), total nitrogen (N), soil pH, clay and sand content
For Bruker, DRIFT spectra ranged from 0.45 to 1.58 and directional hemispherical reflectance (DHR) spectra from 0.28 to 1.46; Agilent spectra showed the greatest minimum to maximum differences (Agilent #4: 0.20 to 1.62)
Summary
Soil spectroscopy has been established as an efficient method to complement conventional soil analysis [1,2,3]. Since estimations of soil parameters with spectral data are based on multivariate calibrations, a large number of statistical modelling approaches that relate the soil information in the spectra to reference values acquired with traditional laboratory methods were evaluated in parallel (see [6,7]). These include—in addition to partial least squares (PLS) regression as the most commonly used method [8]—multivariate adaptive regression splines, artificial neural networks, support vector machines, or regression trees and random forests [9,10,11,12]. Pooled from Agilent #1, Agilent #2 and Agilent #3; corresponds to 2.5 to 15.385 μm
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
