Estratégias de avaliação de experimentos de eucalipto no Vale do Jari na Amazônia

Abstract

The forest sector has several challenges to overcome in the global and national perspectives, such as the demand for forest products due to the increasing in global population and the effect of climate changes, which cause uncertainty on planning and bring insecurity to the forest business. In this context, improving the techniques for selecting superior individuals is strategic, and can be done by using more powerful and accurate models/approaches. Thus, the goal of this study was to assess the efficiency of model mixed use on genetic parameter predictions and to study the accuracy of genetic values for eucalyptus clones using different residual covariance matrixes in trials with different environments using longitudinal data. Data from a clonal test belonging to the breeding program of the company Jari was used, belonging to three sites with 90 hybrid clones with the following species: Eucalyptus platyphylla; Eucalyptus grandis; Eucalyptus urophylla; Eucalyptus wetarensis; Eucalyptus tereticornis; Eucalyptus camaldulensis and Eucalyptus globulus. The trial was assessed twice, at three and six years of age, evaluating three variables: girth at breast height, height and mean annual increment in wood volume. Data was analyzed by the mixed model approach and the model selection was made using the bayesian information criterion (BIC). For all evaluated variables, the mixed model was the best approach, since it caused the reduction of error of prediction and better covariance structure for the dataset. According to BIC, among the three tested residual covariance structures, the best covariance structure was the composed symmetry (CS), which assumes that variances are homogeneous and covariances are constant.