Effect of the sample measurement representativeness on soil carbon determination using near-infrared compact spectrophotometers

Abstract

Compact or miniaturized near-infrared (NIR) spectrophotometers are inexpensive instruments that can boost the use of NIR spectroscopy (NIRS) for determining soil organic carbon (SOC). Nevertheless, these instruments are deficient concerning the probing capability due the small area of the sample assessed in spectra measurement. Protocols normally used during spectral data acquisition can be insufficient to obtain representative spectra, which can affect the performance of the models to estimate SOC. This paper aimed to understand how the representativeness of soil spectra is affected by increasing the area probed during spectra acquisition and determine its effect on the performance of multivariate regression models constructed using compact NIR spectrophotometers for SOC determination. Four measurement modes of the flask containing the soil sample were evaluated using a compact FT-NIR. Measurement modes I (full radial and axial scanning of sample flask) and II (full radial scanning in a fixed axial line of sample flask) were performed using a rotational-linear sample probing device to move the sample flask over the optical probing beam during spectral data acquisition. Measurement modes III and IV, in turn, were performed in static mode. Using the measurement mode III, the spectral data were acquired at variable points of the sample flask, and using the measurement mode IV, the spectral data were acquired at a single fixed point in the sample flask. The study was developed in two stages. Firstly, five spectra of three soil samples (selected according to their carbon content) were acquired using the measurement modes. The repeatability and representativeness of the spectral data were then evaluated using principal component analysis (PCA). Partial least square regression (PLSR) models were constructed to determine the quantitative effects of the representativeness of the four measurement modes used to acquire the spectral data set. The calibration and independent validation sets were composed of 49 and 20 soil samples, respectively. The increase in the sample area probed using the a rotational-linear sample probing device resulted in more representative and reproducible spectral data, which improves the performance of compact spectrophotometer in SOC determination. Significant enhancements in SOC determination ranged from 146 to 400 % were found, considering the most (mode I) over less representative measurement modes. These results were verified by a comparative randomization test between models’ prediction performances. In this context, using a rotational-linear sample probing device is recommended for SOC determinations using compact NIR spectrophotometers.