Abordagem bayesiana para o processo espaço-temporal log gaussiano de Cox com aplicação no setor florestal

Abstract

Through the analysis of Poisson processes has been possible to perform satisfactorily some studies with data point processes counting. However, these processes are limited to the study of situations with homogeneous patterns, hardly found in actual data. This research has proposed the study of Log Gaussian Cox Processes, process that makes possible the study of patterns points heterogeneous data, with a based from Poisson process with on the realization of a Gaussian random field. We did two applications for the process, the first with simulated data of outbreaks of fire in Castilla-La Mancha, Kingdom of Spain, in order to explore the properties of the graph and computational of LGCP, and study the heterogeneity proposed by the process. The second focuses on real data of fire points and average rainfal in the Amazon Biome, Brazil, detected by satellite NOAA 15, between the years 2007 and 2011. The Inference for these processes are carried out under the Bayesian approach, using the Monte Carlo Markov Chain (MCMC). The proposed objectives of this work were completed satisfactorily, enabling future predictions about the data in the study.