User menu
Although white marlin is not a target large commercial longline tuna boats, it is often a bycacth. There is little information on stock structure but it is assumed that the one unique stock in the Indic Ocean is the most probable hypothesis. The available data is limited to catch and catch rates. Usually the quality of the data concerning bycatch species is not high, hence it is difficult to achieve success running stock assessment models. In this paper a potentially useful Bayesian version of state-space biomass dynamic models (Fox and Schaefer types) are used in an attempt to assess the status of the white marlin stock of the Indic Ocean. Results are compared to conventional versions in which only the observational error is considered. Calculations were based on estimations of total catch and on standardized catch rates as estimated based on Japan database. In this and in his companion paper (IOTC2013- WPB11-25) the likelihood function was based on log-normal density distributions. Monte Carlo Markov Chains are used to calculate the posterior sample. Three chains starting with different parameters estimations were calculated. The first 50000 samples of each chain were discarded (burnin), and the next 50000 samples were sliced resulting in a final sample with size equal to 1000. Convergence of the chains was assessed using Gelman-Rubin diagnostics. Most of state-space models have converged, but not the observational error models. The exception among the observational error models was the Fox type as calculated with a non- informative prior. The state-space models are not biased, but the observational error are. The striped marlin database is not very informative, hence the uncertainties on the estimations were very high. The state-space model estimations were also very sensitive to choices about priors. More investigation is needed on the behavior of state-space models when the data is not that informative.
