Espectroscopia de infravermelho próximo em análises de solos e plantas

• Autor(a) principal: Santos, Ana Paula dos
• Data de Publicação: 2011
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFU
• Texto Completo: https://repositorio.ufu.br/handle/123456789/12146
• topic: Calibração multivariada; Near-infrared; Métodos de análises; Solos - Análise; Análise espectral; Calibração; Multivariate calibration; Near-infrared; Methods of analysis; CNPQ::CIENCIAS AGRARIAS::AGRONOMIA

Analyses of soils and plants, essential for crop fertilization management, are based on methodologies that are time and reagent consuming. Near infrared spectroscopy (NIR - Near-infrared) has proven as a faster and cleaner alternative for simultaneous quantification of compounds. This study evaluates the methodologies based on NIR combined with multivariate calibration methods to estimate levels of silicon (Si), organic matter (O.M.) and clay in soils, and total nitrogen (N-Total) in maize and soybean leaves. Calibration of models for soil analysis was done with 170 soil samples while that for leaf analysis used 109 (58 maize and 51 soybean), obtaining their spectra in a near infrared spectroscopy NIR model 900PLS belonging to the Laboratory Fertilizers, Federal University of Uberlandia (Lafer), Uberlândia-MG, where the models were built. The mathematical method used was the partial least squares: PLS. The model validation has been crossed and the number of latent variables ranged between 5 and 8. The reference methods for comparison were: colorimetric-extraction with calcium chloride (0,01 mol L-1) for analysis of soluble Si in the soil, Walkley-Black Colorimetric analysis of O.M. and the pipette method for clay analysis. The models for determination of N-Total were compared to the Semimicro-Kjeldahl method. After calibration and cross validation, the models were tested and the results evaluated using the correlation coefficient (r), the calibration (RMSEC) and the prediction errors (RMSEP), and relative error (ER). Also, Student s t test at 0,05 significance was used to test for equality between averages. Correlations of 0,71, 0,84 and 0,85 were obtained in the calibration of models for analysis of Si content in the soil, O.M. and clay, respectively. The prediction from these models showed low correlations (lower than 0,5) with significant t test for silicon and clay. The RMSEC obtained from Si analysis was 2,03, RMSEP of 5,74 and ER of 59,2%. To O.M. the RMSEC was 0,81, RMSEP = 1,03 and ER = 39,3%. The analysis of clay presented RMSEC = 11,2, RMSEP = 16,55 and ER = 34,7%. In the analysis of maize and soybean leaf nitrogen calibration correlations of 0,85 and 0,88 were obtained, respectively. Good correlations were obtained for prediction of maize (r = 0,80) and soybean (0,76) samples, with non significant t-test, indicating that leaf analysis for the values predicted by the models NIR/PLS did not differ from those of the reference methods. Calibration and prediction errors for N in maize and soybean were smaller than 5,0 g kg-1, and the ER for N analysis in maize (8,3%) greater than that of soybean (5,7%). The observed results demonstrate greater efficiency in the use of near infrared analysis for leaf than for soil analysis.