Espectroscopia no infravermelho próximo para estimativa dos teores de argila e de matéria orgânica em amostras de solo

• Ano de defesa: 2019
• Autor(a) principal: Lazzaretti, Bruno Pedro
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Santa Maria; Brasil; Agronomia; UFSM; Programa de Pós-Graduação em Ciência do Solo; Centro de Ciências Rurais
• Palavras-chave em Português: Calibração; Validação; Modelos matemáticos; Pré-tratamento espectral; Calibration; Validation; Mathematical models; Spectral pretreatment; CNPQ::CIENCIAS AGRARIAS::AGRONOMIA::CIENCIA DO SOLO
• Link de acesso: http://repositorio.ufsm.br/handle/1/19211

Among the constituents of the soil, special attention is given to soil clay and soil organic matter (SOM), since, among other aspects, they are determinant for nutrient retention and for the formation of aggregates, which directly affect the productive potential of crops. The most commonly used methods for quantification of these constituents present some disadvantages, such as the use of chemical reagents and the generation of residues. The Near Infrared Spectroscopy (NIRS) arises as an alternative to such methods. The objective of this work is to develop models for the quantification of clay and organic matter contents in soil samples using spectral data obtained via NIRS. 400 soil samples from the UFSM routine laboratory were used for generating the calibration curve, 100 for each soil clay class (class 1 clay > 60%; class 2 clay between 41 and 60%; class 3 clay between 21 and 40%; and class 4 clay ≤ 20. Clay and organic matter contents were determined via densimeter and sulfochromic solution methods, respectively. The untreated spectra (absorbance) and the pretreated spectra (Savitzky-Golay derivative) of the 400 samples were used for calibration purposes with previously known mathematical models. For calibration, we used models with four algorithms: Multiple Linear Regression (MLR), Partial Last Squares Regression (PLSR), Support Vector Machine (SVM), and Gaussian Process Regression (GPR). The validation of the curve was performed with the model that presented the best performance in the calibration (higher R2 and lower RMSE) and in two ways: with 40 random samples (10 of each clay class) used in the calibration; and with 200 new unknown samples (50 of each class of clay) from the UFSM routine laboratory. The clay content of the soil samples affects the predictive capacity of the calibration curve for the estimation of the SOM content via NIRS. The validation of the curves presented worse performance (higher R² and lower RMSE) when carried out on unknown samples. The model tends to overestimate the lower contents and to underestimate the higher contents of clay and SOM. Despite its potential, in order to use the prediction of these attributes via NIRS in soil analysis laboratories, further calibration studies are necessary.