Performance Comparison of Three Data-Poor Methods With Various Types of Data on Assessing Southern Atlantic Albacore Fishery

Abstract

In the world, more than 80% of the fisheries by numbers and about half of the catches have not been formally analyzed and evaluated due to limited data. It has led to the fast growth of data-poor evaluation methods. There have been various studies carried out on the comparative performance of data-poor and data-moderate methods in evaluating fishery exploitation status. However, most studies to date have focused on coastal fish stocks with simple data sources. It is important to pay attention to high sea fisheries because they are exploited by multiple countries, fishing gears and data may be divrsified and inconsistent. Furthermore, a comparison of the performance of catch-based, length-based, and abundance-based methods to estimate fishery status is needed. This study is the first attempt to apply catch-based, length-based, and abundance-based data-poor methods to stock assessment for an oceanic tuna fishery and to compare the performance with a data-moderate model. Results showed that the three data-poor methods with various types of data did not produce an entirely consistent stock status of the southern Atlantic albacore (Thunnus alalunga) fishery in 2005, as the estimated B2005/BMSY ranged from 0.688 to 1.3 and F2005/FMSY ranged from 0.708 to 1.6. The Monte Carlo Catch maximum sustainable yield model (CMSY) produced a similar time series of B/BMSY and F/FMSY and stock status (recovering) to the Bayesian state-space Schaefer model (BSM). The abundance-based method (AMSY) gave the most conservative condition (overfished) of this fishery. Sensitivity analysis showed the results of the length-based Bayesian biomass estimation method (LBB) are sensitive to Linf settings, and the results with higher Linf were similar to those of other models. However, the LBB results with setting Linf at lower levels produced more optimistic conditions (healthy). Our results highlight that attention should be paid to the settings of model parameter priors and different trends implied in various types of data when using these data-poor methods.