Avaliação visual em braquiária: número de avaliadores e modelos de análise

• Ano de defesa: 2017
• Autor(a) principal: Fonseca, Jales Mendes Oliveira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Lavras; Programa de Pós-Graduação em Genética e Melhoramento de Plantas; UFLA; brasil; Departamento de Biologia
• Palavras-chave em Português: Brachiaria - Avaliação visual; Acurácia seletiva; Forrageiras - Melhoramento genético; Modelo linear misto generalizado; Transformação Box-Cox; Inferência bayesiana; Brachiaria - Visual assessment; Selective accuracy; Forage - Genetic improvement; Generalized linear mixed model; Box-Cox transformation; Bayesian Inference; Melhoramento Vegetal; Genética
• Link de acesso: http://repositorio.ufla.br/jspui/handle/1/12639

Urochloa ruziziensis (R. Germ. & CM Evrard) Crins. (sin. Brachiaria ruziziensis) breeding programs have adopted visual sensitivity as an evaluation parameter of genotypes. Due to the lack of knowledge regarding the adequate number of evaluators in the field and the way of performing the analysis of grades attributed to genotypes, the objective of this study was to determine an optimized number of evaluators and present different alternatives for analyzing grades. In order to do that, 2204 U. ruziziensis genotypes were evaluated in augmented block design using two controls: Marandu (U. brizantha) and Basilisk (U. decumbens). Green mass production was measured and six evaluator attributed vigor grades based on visual inspection. To determine the number of evaluators, data were analyzed using mixed models approach with recovery of interblock and intergenotypic information considering grades given by one evaluator and by the average of 2, 3, 4, 5 and 6 evaluator combinations. In order to present the different alternatives for grade analysis, it was used approaches based on linear mixed models in original score scale, linear mixed models in BoxCox transformed scale, generalized linear mixed models and Bayesian generalized linear mixed models. To validate the analysis, both cases considered genetic and experimental parameters used in breeding programs assisted by visual evaluation. Based on the results, it was possible to conclude that indirect selection of U. ruziziensis genotypes using grades is valid and that the optimal number of evaluators to carry out visual evaluations is three. Moreover, it was not possible to detect significant differences between the alternatives grade analysis, but it was possible to observe the adequacy offered by the usage of models that consider original grade scale.