Search

We used mid‐infrared (MIR) spectra (4000–600 cm−1) to identify and classify soil orders and soil horizons from 102 pedons across five soil orders (Alfisols, Entisols, Mollisols, Spodosols, and Histosols). The soils were analyzed for texture, total carbon, pH, and elemental properties. Random forest models were used to group the spectra of master horizons (O, A, E, B, and C), B horizons (Bs, Bt, and Bw), and the five soil orders. The prediction accuracies for the master horizons and B horizons were 0.81 and 0.89, respectively. The Kappa coefficient was 0.71 for the prediction of master horizons and 0.73 for the prediction of B horizons. The soil orders had an overall accuracy of 0.73 and a Kappa coefficient of 0.64. Histosols exhibited unique absorption characteristics at 2930 and 2860 cm−1 that differed distinctly from mineral soils. The MIR spectra accurately distinguished the O horizons. The spectral curve of topsoil of Spodosols was comparable to the O horizons. Spodosols under forest had A horizons with high organic matter and were classified accurately. Entisols (Psamments) displayed absorption peaks associated with sand, facilitating their differentiation from the other soil orders. The model struggled to discern subtle differences among some soil orders, and identification is hampered if soils undergo irreversible changes upon drying. However, our results showed that MIR spectra can be used for effectively identifying and classifying soil orders as well as soil horizons. Core Ideas: Mid‐infrared (MIR) spectra measured from 102 pedons across nine regions in Wisconsin.Unique adsorption characteristics in MIR spectra reflected pedogenetic features and horizons.MIR spectra and random forest machine learning correctly identified soil horizons and orders.Master horizons were classified with an accuracy of 0.81 and B horizons with an accuracy of 0.89.Soil orders were classified with an accuracy of 0.73. [ABSTRACT FROM AUTHOR]