User menu
To grasp the historical trajectory of Swordfish stock abundance, we addressed standardizing the CPUE of Swordfish in the Indian Ocean by Japanese longliners using their logbook data for the period 1979-2018. We divided the time-period into two periods, 1979-1993 and 1994-2018 for the analysis for four areas (NW, NE, SW, SE) of Indian Ocean because of apparent change of data-format of logbook around in 1994 and the change of fishing methods (e.g. materials of stem and branch lines and gear configuration such as number of hooks between floats) related to catchability: q not detailed in the logbook during the mid-1990s. In this analysis, we applied Bayesian hierarchical spatial models. Since the catch data are counts characterize by many zeros, we evaluated zero-inflated Poisson GLMM (ZIP-GLMM). All analyses were performed using R, specifically the R-INLA package. The INLA procedure, in accordance with the Bayesian approach, calculates the marginal posterior distribution of all random effects and parameters involved in the model. We applied half Cauchy distribution as a prior for the random effect. Best candidate models were selected based on Widely Applicable Bayesian Information Criterion (WAIC). From the lowest value of WAIC, spatial Poisson GLMM with autoregressive (AR1) modelled for the year trend (i.e. m_zip_spde2 model) was selected as the best candidate for each area except for SE area. No apparent trend in interannual variation of standardized CPUE was generally observed for each area. The uncertainties are much larger for the current spatial models due to consideration of spatial effect as compared to the past non-spatial models (Ijima 2017) although the trend of point estimates is similar. We will improve the models dealing with the appropriate catchability (q), applying the state space model and/or latent variable model in the future.
