Mineral and particle-size chemometric classification using handheld near-infrared instruments for soil in Northeast Brazil

Abstract

The characterization of soil variations crucial for agriculture is challenging due to soil having different mineral composition and particle-size distribution. Traditional methods are costly and time-consuming for large-sized areas. Spectroscopic techniques coupled with chemometrics are alternative ways to overcome these drawbacks. Miniaturized near-infrared (NIR) spectrophotometers provide fast, cost-effective spectra acquisition for assessing soil chemistry and distribution despite challenges like overlapped bands and reduced spectral range. This study presents a pattern recognition strategy to address these limitations, enhancing the use of handheld NIR instruments for soil analysis. The study analyzed 176 soil samples from 15 soil groups in the Northeast region of Brazil. First, attenuated total reflectance-Fourier transform infrared (ATR-FTIR) and energy-dispersive X-ray fluorescence spectrometry (EDXRF) were employed to characterize the samples, providing complementary vibrational and elemental information, respectively. Common Dimension Analysis (ComDim) identified links between ATR-FTIR and EDXRF data, aiding soil characterization. The Common Components (CCs) from ComDim were used in Partitioning Around Medoids (PAM) clustering, resulting in five distinct classes based on their mineral composition. These classes showed significant differences in clay and sand contents. With the use of ComDim-PAM, samples were labeled for classification via Partial-least Squares-Discriminant Analysis (PLS-DA) and Support Vector Machine (SVM) using NIR spectra (spectral range: 908–1676 nm) from two handheld instruments (Hand 1 and Hand 2). The SVM models outperformed the PLS-DA models, especially by including variable selection for Hand 1, with test accuracy exceeding 90%. These findings highlight the method's advantages for fast and cost-effective assessment, classification, and soil mapping based on their mineral and particle-size distribution.