Modelo espacial birnbaum-saunders aplicado a dados agrícolas

Abstract

Understanding the spatial distribution knowledge regarding georeferenced data has been essencial to various areas including agriculture. Thus, several trials have been carried out. However, most of these studies assume that the underlying stochastic process is Gaussian. When the data associated with this process do not present normality, data transformations are applied. And though the use of these transformations has presented satisfactory results, it is important to consider models which take into account the characteristics of such phenomenon. It may be more appropriate than using a normal model. So, this trial aimed at proposing a spatial model based on the Birnbaum-Saunders distribution (BS). This distribution has been shown effective to model data that take positive values and whose behavior presents positive asymmetry and unimodality. Thefore, this trial has proposed a methodology that includes the formulation of the spatial Birnbaum-Saunders model , estimation of its parameters using maximum likelihood (ML), and application of diagnostic techniques which can detect the sensitivity of the model to atypical data and evaluate the proposed model through a simulation study and studies using real data sets of agricultural engineering. These data were obtadined in a 167.35-ha commercial area for grain production, in Cascavel city, to validate the studied model. In the study with simulated data and large samples, estimation parameters and diagnostic analysis showed a good performance. According to the study with real data, calculations of AIC (Akaike s information criterion) and BIC (Bayesian information criterion) indexes, Bayes factor as well as Q-Q plots constrution have shown that the proposed model is appropriate to fit the obtained data. Influential cases were detected, and their removal from data set caused a considerable change in contour maps. It is therefore concluided that Birnbaum-Saunders spatial model is adequate to carry out studies with spatially correlated data. Is is also an alternative model to the normal model when the data set present positive asymmetrical distribution