The potential of mid- and near-infrared diffuse reflectance spectroscopy for determining major- and trace-element concentrations in soils from a geochemical survey of North America

In 2004, soils were collected at 220 sites along two transects across the USA and Canada as a pilot study for a planned soil geochemical survey of North America (North American Soil Geochemical Landscapes Project). The objective of the current study was to examine the potential of diffuse reflectance (DR) Fourier Transform (FT) mid-infrared (mid-IR) and near-infrared (NIRS) spectroscopy to reduce the need for conventional analysis for the determination of major and trace elements in such continental-scale surveys. Soil samples (n = 720) were collected from two transects (east-west across the USA, and north-south from Manitoba, Canada to El Paso, Texas (USA), n = 453 and 267, respectively). The samples came from 19 USA states and the province of Manitoba in Canada. They represented 31 types of land use (e.g., national forest, rangeland, etc.), and 123 different land covers (e.g., soybeans, oak forest, etc.). The samples represented a combination of depth-based sampling (0-5 cm) and horizon-based sampling (O, A and C horizons) with 123 different depths identified. The set was very diverse with few samples similar in land use, land cover, etc. All samples were analyzed by conventional means for the near-total concentration of 49 analytes (C_total, C_carbonate and C_organic, and 46 major and trace elements). Spectra were obtained using dried, ground samples using a Digilab FTS-7000 FT spectrometer in the mid- (4000-400 cm^-1) and near-infrared (10,000-4000 cm^-1) at 4 cm^-1 resolution (64 co-added scans per spectrum) using a Pike AutoDIFF DR autosampler. Partial least squares calibrations were develop using: (1) all samples as a calibration set; (2) samples evenly divided into calibration and validation sets based on spectral diversity; and (3) samples divided to have matching analyte concentrations in calibration and validation sets. In general, results supported the conclusion that neither mid-IR nor NIRS would be particularly useful in reducing the need for conventional analysis of soils from this continental-scale geochemical survey. The extreme sample diversity, likely caused by the widely varied parent material, land use at the site of collection (e.g., grazing, recreation, agriculture, etc.), and climate resulted in poor calibrations even for C_total, C_organic and C_carbonate. The results indicated potential for mid-IR and NIRS to differentiate soils containing high concentrations (>100 mg/kg) of some metals (e.g., Co, Cr, Ni) from low-level samples (<50 mg/kg). However, because of the small number of high-level samples, it is possible that differentiation was based on factors other than metal concentration. Results for Mg and Sr were good, but results for other metals examined were fair to poor, at best. In essence, it appears that the great variation in chemical and physical properties seen in soils from this continental-scale survey resulted in each sample being virtually unique. Thus, suitable spectroscopic calibrations were generally not possible.