Full text
| |
| Author(s): |
Maria Del Pilar Diaz
Total Authors: 1
|
| Document type: | Doctoral Thesis |
| Press: | Piracicaba. |
| Institution: | Universidade de São Paulo (USP). Escola Superior de Agricultura Luiz de Queiroz (ESALA/BC) |
| Defense date: | 1996-09-03 |
| Advisor: | Hilton Thadeu Zarate do Couto |
| Abstract | |
In the forestry area, a tree mortality estimation for a given species must be present in the reliable prediction systems for the species growth. However, the individual mortality study introduces a dichotomic process, arising to count data set for the experimental units. Thus, the identification of data distribution is the previous and important work to know the mortality behavior related to additional information, as the species selection, rotation, age, and site quality. For independent observations mortality, the basic Poisson model could be proposed, nevertheless it is frequently to find that the sample variance is greater than the nominal one, which is called overdispersion. The present work identifies the probabilistic model of Eucalyptus grandis through formal and informal technics. As the genus Eucalyptus is very important in the forestry brazilian production, it was chosen. The E. grandis, belonging to the genus, has desirable properties, like the environment adaptation, fast growth, so that it becomes one of the most useful species for the paper and cellulose national production. Under generalized linear model approach, a sistematic component with two factors, rotation species and site quality. The aim was to study two methodologies for the overdispersion Poisson models, that is, the Extra-Poisson model, through maximum quase-likelihood, and the Negative Binomial model, through maximum likelihood method. The dispersion parameter was estimated by 2.18, which means that the mortality distribution has a hight agregated index. Due to the fact that the maximum likelihood algorithm is complex, simple alternative methods were suggested. Their performances related to the efficiency were studied too. The results indicated that the proposed maximum likelihood method might be used. Finally, a discret exponential model was shown like a greater theoretical support to generalize the model keeping the true probabilistic data structure. This generalization establishes the background to formulate a stocastic process for the tree mortality following the changes as the time goes by. (AU) | |