Determinação de parâmetros de fertilidade do solo por meio da análise multivariada de imagens e de espectros de infravermelho

• Ano de defesa: 2020
• Autor(a) principal: Morais, Pedro Augusto de Oliveira
• Orientador(a): Oliveira, Anselmo Elcana de
• Banca de defesa: Oliveira, Anselmo Elcana de, Madari, Beata Emoke, Antoniosi Filho, Nelson Roberto, Coelho, Clarimar José, Souza, Aparecido Ribeiro de
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Goiás
• Programa de Pós-Graduação: Programa de Pós-graduação em Química (IQ)
• Departamento: Instituto de Química - IQ (RG)
• País: Brasil
• Palavras-chave em Português: Textura do solo; Carbono orgânico do solo; Óxidos; MIA; MIR; PLS; SPA-MLR; LS-SVM
• Palavras-chave em Inglês: Soil texture; Soil organic carbon; Oxides; MIA; MIR; PLS; SPA-MLR; LS-SVM
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::QUIMICA
• Link de acesso: http://repositorio.bc.ufg.br/tede/handle/tede/11503

Soil analysis is an important tool when monitoring the environmental impact of agricultural activity. It also allows for the rational planning inputs contributing to a better environmental sustainability and economic production. Consequently, there is a growing demand for the services of soil analysis laboratories. However, methodologies currently employed in the field not only generate a considerable amount of waste, but also have a high set up cost. Therefore, cheaper and environmentally sustainable alternatives should be developed. In this sense, this study proposes the use of soil digital images and mid-infrared spectroscopy (MIR) to estimate soil organic carbon (SOC), predict and classify soil texture, as well estimate iron, aluminium, and silicon oxides contents. For this purpose, 177 samples from different regions of the country were analyzed by standard methods. Soil digital images were acquired using RGB (Red, Green, Blue) in Tiff format. The correlation between digital images, MIR spectrum, and soil fertility parameters was obtained using Partial Least Squares Regression (PLS), Multiple Linear Regression algorithm associated with the Successive Projections (SPA-MLR), and Least Squares Support Vector Machines (LS-SVM). The best models present correlations higher than 90% and Residual Prediction Deviation (RPD) values greater than 3.0. The use of these methods in test soil analysis would allow a significant increase in productivity, reduction of the cost of analysis, and minimization of environmental impact. The propsed analyses do not produce waste and do not employ chemicals. As a result, farmers can benefit from the proposed methods taken into account that the analyses are quick and inexpensive and might lead to an increase in productivity in the field.