I made two sensitive runs of CASAL catch at length models (LENGTH_MODEL) for stock status assessments of Dissostichus eleginoides in research block 5843a_1 in Division 58.4.3a for the years 2004/05 to 2012/13 following the recommendation during the last WG-SAM meeting. The two models are generally based on those with the same names in WG-SAM-14/17 as follows: 1) R.0.1 model without tagging events before 2008, and 2) R.1 model including all tagging events (years 2005-2012). These models were comprised of two fisheries split by depths of 1 200 m as the model in WG-SAM-14/17. Number of longlines was used as (unknown) multinomial sample size for each model. The over-dispersion for the tag-recapture likelihoods was estimated using a mean-based weighting method for each model. The other common conditions between the two models were set as follows: year class strength (YCS) was fixed; parameters of ALK were fixed; parameters of selectivity for legal fishery were estimated; uniform prior was used for initial biomass (B0) estimate.
The MPD estimate of B0 was 464 and 772 tonnes in R.0.1 and R.1 models, respectively. The current vulnerable biomass was 647 and 1 201 tonnes, respectively. The median MCMC estimates for the biomasses (490 and 810 tonnes) were slightly higher than those by MPD for each model. MCMC posterior traces for B0 seem to be rather convergent for each model.
Further sensitive runs for R.1 model (R.1_DER_MODEL) were examined using six options changing conditions, i.e. estimate/fix for given parameters, uniform-log prior for B0, and multinomial-based weighting for tag-recapture. The difference of median estimate of B0 between MPD and MCMC and the range of MCMC were considerably large for R.1_2 model, which used multinomial-based weighting for tag-recapture that lead to quite high value of dispersion.