O modelo Log-Poly-Weibull multivariado aplicado à predição espaço-temporal de temperatura no Estado do Amazonas

Abstract

A class of multivariate distributions for possibly correlated data was built by mar-ginalizing of G-exponentialized Type IV Pareto distribution, and conditioned to a mixture of positive alpha-stables distributions. Some proprieties of this class were presented and the procedures for estimation and inference were discussed. A Log-Poly-Weibull spatial model was introduced, as a special case of proposed class, for modeling and space-temporal prediction of temperature surfaces. Most properties of the model were similar to the ge-neral model ones, while some were added to fit with the application. The parameters were estimated by the MCEM algorithm and the standard error through parametric method Bootstrap. The proposed model was applied to data on mean compensated tempera-ture observed from 1996 to 2015 at meteorological stations of Amazon State — Brazil and simulations were performed to evaluate the predictive capacity of the model.