Proposição de níveis críticos e predição de nutrientes por espectroscopia VIS-NIR em folhas de pessegueiros

• Ano de defesa: 2024
• Autor(a) principal: Hindersmann, Jacson
• 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: Prunus persica; Valores de referência; Linha de fronteira; Aprendizado de máquina; Biblioteca espectral de folhas; Reference values; Borderline; Machine learning; Leaf spectral library; CNPQ::CIENCIAS AGRARIAS::AGRONOMIA::CIENCIA DO SOLO
• Link de acesso: http://repositorio.ufsm.br/handle/1/32001

The peach tree (Prunus persica L. Batsch) is a fruit tree of great economic importance in the world. However, the average crop productivity in Brazil is far below that obtained in other traditional producing countries. The lower Brazilian productivity may be related to nutritional problems, since the reference values currently used, such as critical levels (CL) and nutrient sufficiency ranges (SR), were established in a general way, that is, they were not obtained for specific regions and cultivars. In addition, thousands of chemical leaf analyzes are carried out annually to help make decisions about the application of fertilizers. However, traditional methods used to determine the concentration of nutrients in tissue require the use of a mixture of strong acids, in addition to being a time-consuming analysis. Visible (Vis) and near infrared (NIR) spectroscopy techniques emerge as a possible solution to overcome the limitations presented by traditional chemical analyses. The studies in this work aimed to propose nutritional reference values and estimate the concentration of nutrients by Vis-NIR spectroscopy in leaves of peach trees grown in southern Brazil. Study 1 proposed the CL and SR reference values in leaves of two peach cultivars, 'PS10711' and 'Maciel', using the Bayesian Segmented Quantile Regression Frontier Line (RQSB) method for the nutrients N, P, K, Ca, Mg, S, B, Cu and Fe. A database was used with productivity and nutrient concentrations in leaves, obtained in two harvests in the Pinto Bandeira region and in three harvests in the Pelotas region. Study 2 estimated the concentration of nutrients (N, P, K, Ca, Mg, S, B, Cu, Fe, Mn and Zn) in leaves of peach trees cultivated in the two aforementioned regions using spectral data obtained with the Vis technique. -NIR combined with the partial least squares regression (PLSR) machine learning method with pre-processed data with the 1st derivative of Savitski-Golay (SGD1d). In Study 1, RQSB was efficient in proposing new CLs for the nutrients N = 31 g kg-1, K = 23.5 g kg-1, P 'PS' = 2.8 g kg-1 and P 'Maciel' = 2.0 g kg-1, Ca 'PS' = 28.5 g kg-1 and Ca 'Maciel' = 19.5 g kg-1, Mg 'PS' = 4.5 g kg-1 and Mg ' Maciel' = 6.0 g kg-1, S 'PS' = 0.85 g kg-1, B 'PS' = 25 mg kg-1 and B 'Maciel' = 35 mg kg-1, Cu 'PS' = 9.0 mg kg-1 and Cu 'Maciel' = 6.3 mg kg-1, Fe 'PS' = 72 mg kg-1 and Fe 'Maciel' = 56 mg kg-1. In study 2, the local-1 ‘PB’ model showed higher accuracies in nutrient prediction compared to the regional model ‘PB+Pelotas’ and the local-2 model ‘Pelotas’. The results observed in Studies 1 and 2 reinforce the need for new studies with the themes addressed here.