Aplicações da biometria em programas de melhoramento do eucalyptus

Abstract

The Brazilian forestry sector represents a significant part of the national economy, employing thousands of people both directly and indirectly. Eucalyptus spp. comprises a substantial portion of planted forests, and its wood is utilized for various purposes. Unlike conventional species such as corn, soybeans, beans, and others, forest species have not undergone any type of breeding in the past. Thus, the potential for genetic improvement of forest species is high. The genotype by environment interaction (GxE) directly affects the performance of Eucalyptus spp. clones planted in different Brazilian regions, making it an important area of study in plant breeding. Therefore, the first chapter of this study is dedicated to investigating the GxE interaction of 62 Eucalyptus spp. clones, as well as their adaptability and stability. The clones were evaluated over 10 years in 10 sub-regions in the states of São Paulo and Minas Gerais. The collected data on mean annual increment (MAI) were subjected to the methodologies of Annicchiarico (1992), Eberhart & Russel (1966), and GGE biplot. Regions H, I, G, J, and D exhibited the best average performances, respectively. Considering the GxE interaction, the best clones within the best sub-regions were: H-1112, I-887, G-1112, J-838, D-107. Clones 876, 1112, 1277, and 13 showed the highest stability, while clones 1112 and 1277 presented not only high stability but also the highest average performance and adaptability. The second chapter of this study aimed to compare, through stochastic simulation, two breeding methodologies: intrapopulation recurrent selection (IRS) and interpopulation recurrent selection (IRR). These two cyclic methodologies are frequently used in eucalyptus breeding, but due to the high demand for time and labor, the practical comparison of these two methods is unfeasible. Thus, the use of computational simulations can assist in comparing these two methodologies for longer cycles, which would be unfeasible in practice. The simulations were performed using the AlphaSim package in the R® software. Twenty-four selection cycles were simulated, totaling 250 years, with additive variance (Va) values equal to 5, 10, 20, and 30, in combination with degree of dominance (DD) values equal to 0, 0.3, 0.7, and 1. At the end of the simulations, it was found that IRS yields higher gains in scenarios where DD is equal to zero, similar gains to IRR when DD is equal to 0.3, and lower gains when DD is equal to 0.7 and 1.