Estimativa da acidez potencial em solos utilizando a espectroscopia Vis-NIR

• Ano de defesa: 2022
• Autor(a) principal: Kunz, Karine Mariele
• 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: Pré-processamento; Modelos matemáticos; Calibração; Validação; Pre-processing; Mathematical models; Calibration; Validation; CNPQ::CIENCIAS AGRARIAS::AGRONOMIA::CIENCIA DO SOLO
• Link de acesso: http://repositorio.ufsm.br/handle/1/28294

The chemical analysis of soil components is a tool that allows good practices for correctives and fertilizers recommendation and managing soil fertility. Traditional methods of analysis usually consume a high number of reagents and require a lot of time for sample preparation and extractions. An alternative has been spectroscopy in the visible and near infrared (Vis-NIR) region. However, this tool needs validation and calibration of models for reliable estimates for different soil parameters. The objective of this work was to evaluate the reliability of Vis-NIR spectroscopy to estimate the potential acidity of tropical soils compared with values obtained by traditional methods used in routine soil analysis laboratories. We used 240 soil samples from agricultural areas and analyzed in the UFSM routine laboratory, 60 samples of each clay class (class 1: clay content ≤ 20; class 2: 21-40; class 3: 41-60; class 4: >60), which are subdivided by organic matter (OM) content into 20 samples of low content class (low ≤ 2.5), 20 samples of the medium class (medium 2.6 - 5.0), 20 samples of the high OM content class (high >5.0). For the validation of the models, 51 unknown samples were used, which were not part of the initial sample bank. The determination of the potential acidity of the samples was made by estimating with the SMP index and by the calcium acetate method. Five spectra pretreatments were used: smoothed (SMO), Savitzky-Golay Derivate (SGD), Multiplicative Scatter Correction (MSC), Continuum Removal (CRR) and Standard normalization variate (SNV). Prediction models for the potential acidity content were developed from raw and pre-processed spectral data. The models tested were Cubist, Multiple Linear Regression (MLR) and Partial Least Squares Regression (PLSR). The evaluation of the precision of the calibration curves was performed using the coefficient of determination (R² ) and the deviations from the root mean square error (RMSE). Curve validation was performed with the model that presented the best calibration performance. Soil spectra showed features related to soil constituents, mainly in SNV, MSC, SGD and CRR techniques. The pre-processing that obtained the best performance in both the calibration and validation stages was the CRR, regardless of the model used. There was a wide variation in the accuracy of the same multivariate method when different preprocesses were applied. The Cubist model presented the best performance, both for validation of samples analyzed by calcium acetate (R²=0.86; r=0.93) and for the SMP index (R²=0.91; r=0.95). Both the calcium acetate method and the SMP index method showed good fit with the model (R²=0.55 and R²=0.53, respectively). Vis-NIR spectroscopy has the potential to estimate the potential acidity, however, other studies and tests are needed to better elucidate the technique until the use of curves in soil analysis laboratories.