Using Bayesian state-space modelling to assess Trichiurus japonicus stock in the East China Sea

ZHANG Kui;CHEN Zuozhi;Key Laboratory of South China Sea Fishery Resources Exploitation & Utilization, Ministry of Agriculture,South China Sea Fisheries Research Institute, Chinese Academy of Fishery Sciences;

Bayesian state-space model;;Trichiurus japonicas;;stock assessment;;Markov Chain Monte Carlo;;biological reference point

Hairtail(Trichiurus japonicus) is one of the most economically important fish species in the East China Sea and supports one of the most valuable fisheries in China. From 1990 to 2012, the total catch for this fishery ranged from 0.39 to 0.91 million tons. However, most studies on this fishery concentrated on feeding habit, variations of catches, trophic composition, and the stock-recruitment relationship. For management, yield per recruit and surplus production models were applied to analyze the data of this fishery and provide a rough MSY estimate of approximately 7.5×105 tons. Until now, reports on the use of stock assessment models for this fishery are limited, and no uncertainty assessment has been undertaken. Therefore, Bayesian state-space modelling was applied to the catch and catch per unit effort(CPUE) data for this fishery. A state-space model describes the dynamics of two related processes: the observation process, which is a function of the unobserved state process, and the state process, which describes the unobserved population dynamics in terms of biomass or abundance. In the present study, the Pella–Tomlinson surplus production model was used for the state process. We used Bayesian inference because it can take into account more uncertainties that are linked to parameters. In this study, four models were constructed based on Markov Chain Monte Carlo simulation with a mix of information and non-information priors. Marginal posterior distributions of model parameters, biological reference points(BRPs), and unobserved variables were based on 250000 iterations after discarding the first 50000 burn-in iterations to ensure no persistent initial pathologic behavior. Results showed that the best-fit of the four models was model 1, with lognormal priors for the intrinsic rate of increase r and carrying capacity K based on deviance information criterion. Gelman & Rubin's method was applied for convergence diagnostics, and WINBUGS software computed the results of the autocorrelation diagnostics. The parameters in model 1 were best fit and passed all the diagnostics. The prior distributions had a significant impact on the results of r and K, which indicates that the data are sensitive to the type of prior distributions of r and K. The significant difference between the prior and posterior distributions of r and K indicate that the data provide more information than the prior distribution for Bayesian analysis. BRP results showed that hairtail stock was overfished from 1995 to 2010(catch over maximum sustainable yield) and faced a serious threat from 2000 to 2006(fishing mortality coefficient over FMSY). The stock was in a good state in 2012 but required persistent management. Because of possible statistical distortion, the results of MSY and BMSY may be overrated. The estimated results from 2004 to 2012 also have uncertainties, because the hairtail fishery in the East China Sea was also influenced by monsoon, precipitation, and other environmental factors.