Since the 70s, spectroscopic techniques, especially those with infrared radiation (NIRS -Near Infrared Spectroscopy and DRIFTS - Diffuse Reflectance Fourier Transformed Spectroscopy) have been proposed for inorganic soil analysis, but only few methods found routine use in Brazil so far. This fact is probably due to the difficulty in building calibration models based on multivariate tools with real soil samples that have a complex composition that varies geographically and according to the management. The objective of this study was to build models for clay, silt and sand fractions of soils from two experimental farms of Embrapa, and assess the influence of the clusters of samples and spectral variable selection on the model quality. For this purpose, reference values of clay, silt and sand were correlated with NIRS and DRIFTS absorbance values by the statistical tool of Partial least-squares (PLS) modeling. High regression coefficients were obtained (0.95, 0.90 and 0.91 for clay, silt and sand, respectively), in the external validation. These results confirm the application of spectroscopic techniques on textural analysis of soil for agricultural purposes. The grouping of samples by location and the selection of spectral variables had little influence on the model quality. The most suitable spectroscopic technique for this purpose was DRIFTS.
soil texture; spectroscopy techniques; infrared radiation; Latosol; clay
