Alternative stock assessments of Pacific saury in the western North Pacific Ocean
This report provides evaluation of the current status of Pacific saury (Cololabis saira) stock in the North Pacific Ocean by employing two surplus production models in continuous time: “A Stock Production model Incorporating Covariates” (ASPIC) BETA Ver. 7.03 (Prager 1994, 2002) and a “stochastic Surplus Production model in Continuous Time” (SPiCT), and one production model in discrete time: COMBI4 (Babayan and Kizner 1988, Babayan et al. 2011). SPiCT is a state-space biomass dynamic model which incorporates seasonality (Pedersen and Berg 2017).
TWG-PSSA conducted stock assessment analysis by employing the Bayesian state-space biomass dynamic models for the period from 1980 to 2015. We successfully fitted other models in different settings: 1) non-Bayesian (Least squares in ASPIC) for longer period from 1950 to 2015; 2) non-Bayesian (log Least squares in COMBI4) for shorter period from 2003 to 2015, and 3) stochastic seasonal state-space biomass dynamic model (SPiCT) from 2000 to 2015.
SPiCT can also account for process and model errors in addition to observation errors in the biomass indices with priors, but we could follow only the third one of the three base-case scenarios, that differed in survey catchability (q) of the Japanese survey biomass index prior: including all CPUEs and free q. Including all CPUEs and q prior defined from 0 to 1 (Base-case scenario-1) was possible in ASPIC, as well as including CPUEs and q prior fixed at 1 (Base-case scenario-2); however, none of the attempts to use Maximum a posteriori (MAP – is a form of penalized likelihood using Bayesian priors) converged with all CPUEs in the input and period of time from 1980 to 2015. The attempts to fix q prior at 1 in SPiCT are not comparable to the results of TWG-PSSA, because fixing was done with highly informative prior (log SD = 0.001), nevertheless estimated q was much higher than 1. Such fixing made SPiCT more robust to initial values.
Two sensitivity analyses were conducted in SPiCT: 1) without the Japanese survey biomass index (excluding survey q) and 2) using only Russian CPUE indices on a monthly scale. In addition, there are many other sensitivity checks in ASPIC and COMBI4 sections.
Estimated parameters by China, Chinese-Taipei, and Japan are given in the Table 8-1 of the Report (NPFC-2017-TWG PSSA01-Final Report). Mean MSY of China ranged from 506.5 to 593.5; that of Chinese-Taipei ranged from 542.3 to 606.7, and that of Japan ranged from 514 to 622 thousand metric tons (MT). We have got MSY in ASPIC ranging from 493 to 685 (Mean = 594) thousand MT; in COMBI4 MSY was 577 thousand MT; and in SPiCT MSY ranged from 429 (highly informative q) to 478 (free q) thousand MT. The highest MSY was estimated in SPiCT without biomass index (1,557 thousand MT), but exclusion of biomass index and Chinese-Taipei CPUE decreased MSY estimate down to 450 (362–560 CI) thousand MT.
We confirm the conclusion of TWG-PSSA that Base-case scenario-3 (free q) indicated a value of q >1 in SPiCT. ASPIC and COMBI4 showed the same B2016/BMSY (>1) and F2015/FMSY (<1) states that were calculated by TWG-PSSA. But SPiCT estimates were not so optimistic with B2016/BMSY = 0.4 (0.24–0.67 CI) and F2015/FMSY =1.07 and a huge uncertainty in Base-case scenario-3 (free q). Exclusion of biomass index provided optimistic states with B2016/BMSY (>1) and F2015/FMSY (<1) in SPiCT, as well as the exclusion of all CPUE indices except for monthly CPUE index calculated by Russia.
Therefore, we suggest that all other members provide seasonal (monthly) estimates of CPUE indices and catches to improve inputs for SPiCT. The use of SPiCT can help us to overcome at least one discrepancy in CPUE indices: multidirectional tendencies, which possibly occur due to the differences in time (and consequently in space) for fishing operations conducted by different members.