Abstract
The objective of this work was to evaluate the use of near-infrared spectroscopy (NIRS) to predict P, C, and N forms in soils developed from basalt, using multivariate models and soil samples with different granulometries. Samples from Oxisols and Alfisols from croplands, pasture, and native vegetation areas were collected at the 0.0-0.20 and 0.60-0.80-m soil layers, and analyzed for contents of available P, remaining P, organic C, total C, and total N. Spectra in the region between 1,100 and 2,500 nm were obtained on samples with granulometry smaller than 2.0 and 0.2 mm. Multivariate prediction models were constructed and validated from the results obtained by the reference methods and spectra. Sample screenings through 0.2 mm mesh improved the relationship between wavelengths in NIRS and the contents of organic C, total C, and total N in the soil, which allowed better calibration and prediction models for these attributes. Irrespectively of particle sizes, the models did not show appropriate robustness, accuracy, and precision for extractable P and remaining P.
Index terms:
Oxisols; Alfisols; partial least squares regression; particle size
