Near-Infrared Reflectance Spectroscopy Prediction of Soil Properties: Effects of Sample Cups and Preparation

Most methods for soil analysis are based on wet chemistry. Near infrared refl ectance spectroscopy (NIRS) is a cost-eff ective and environmentally sound alternative technique. Th is study evaluated the eff ect of sample fi neness (0.2, 0.5, 1, and 2 mm) and sample cups (transport versus spinning) on the accuracy of NIRS predictions of soil texture, cation-exchange capacity, pH, total C and N, organic C, and potentially mineralizable N (N min ) using 150 air-dried samples collected from a 15-ha site dominated by Humaquept, Endoaquept, and Dystrochrept soils. Th e best spectral pretreatment was determined for each property. Principal component analysis (PCA) was used to select samples in calibration and validation sets. Calibration equations were developed using the modifi ed partial least square regression. Th e accuracy of NIRS prediction was evaluated using three statistics for the prediction set: coeffi cient of determination (R 2 ), ratio of performance deviation (RPD), and ratio error range (RER). Across the factorial designed treatments, successful calibrations were observed for clay, sand, and N min (R 2 ≥ 0.90, RPD ≥ 3, RER ≥ 15). Prediction accuracy of pH was poor (0.51 ≤ R 2 ≤ 0.74, 1.39 ≤ RPD ≤ 1.92, 6.13 ≤ RER ≤ 8.33), while it was intermediate for remaining properties. Sample fi neness of 2 mm appeared to be suffi cient since fi nenesses of 0.2, 0.5, or 1.0 mm did not improve calibration accuracy. Th ese fi ndings at small scale should not be extrapolated and further investigations are required to validate them at a larger scale. Abbreviations: b (in italics), slope or regression coeffi cient; CEC, cation-exchange capacity; Corg, organic C; CV, coeffi cient of variation; GH, global Mahalanobis distance (global H); H-outliers, outliers determined using the GH (H-statistic); NH, neighborhood Mahalanobis distance; NIRS, near infrared refl ectance spectroscopy; Nmin, potentially mineralizable N; OM, organic matter; PCA, principal component analysis; PLS, partial least squares; R 2 , coeffi cient of determination; RER, ratio of error range; RPD, ratio