MODELAGEM DA ESTRUTURA TEMPORAL DE CAPTURAS INCIDENTAIS EM PESCARIAS COMERCIAIS ATRAVÉS DE MODELOS HIERÁRQUICOS BAYESIANOS

Abstract

In commercial sheries, catches of species that are not the main target, are called "bycatch". In the Atlantic Ocean, the Black Marlin (Makaira nigricans) is caught incidentally by vessels using longline as shing gear, and targeting tunas and swordsh. Analysis of temporal patterns in a historical context can help in decisions for sheries management. The aim of this work is to build, evaluate and compare models that describe the temporal structure of historical catches of Blue Marlin. Bayesian hierarchical models with latent Gaussian structure were used to model marlin catches in the South Atlantic Ocean, with data from the Japanese eet from 1970 to 2009. Models with random walk (RW) and autoregressive structures were used, as well as a seasonal component for trimesters. Parameters were estimated in a Bayesian context through the Integrated Nested Laplace Approximation (INLA) approach. The negative binomial distribution turned out to be a better alternative for this kind of data, as overdispersion was detected when considering the Poisson distribution. The most appropriate temporal model was autoregressive of order 10. The time trend eect was more pronounced than the effect of seasonality. Inference using the INLA method demonstrated be an effective alternative for use in large databases, such as the one used in this work.