Eveson et al.: Incorporating fishery observer data into an integrated catch-at-age and multiyear tagging model 507 



modeling multiple cohorts simultaneously will give very 

 similar results to modeling each cohort individually. The 

 results would be identical if the catch-at-age data for all 

 cohorts were modeled as independent, but there is likely 

 to be correlation between catch estimates for different 

 cohorts in the same year that should be accounted for. 

 If some constraints can be put on the parameters, such 

 as natural mortality varying only with age or fishing 

 mortality following an age selectivity curve, then 

 precision in the parameter estimates should improve. 

 For example, we re-ran the simulations for scenario 1 

 as described in the model performance section, but with 

 data generated for three cohorts instead of one. In fitting 

 the model, both natural mortality and fishing mortality 

 were allowed to vary only with age. In comparison to 

 the CVs obtained for scenario 1 with one cohort (Table 

 3j, the CVs obtained with three cohorts were roughly 

 35-40% less for all parameters. Again, to determine 

 which parameter constraints are most appropriate in 

 a real situation, standard model selection procedures 

 such as AIC can be used. 



As was illustrated, the BPO model can be used 

 to evaluate the effect of releasing more tags versus 

 increasing observer coverage on the precision and bias 

 of the parameter estimates. Because these programs 

 can be costly to run and resources are usually 

 limited, it is useful to have a statistical framework for 

 comparing how alternate allocations of resources affect 

 the results that can be achieved. Our results confirm 

 the general conclusion of Polacheck and Hearn (2003) 

 that it is important to ensure both adequate numbers 

 of tag releases and adequate observer coverage (the 

 latter for robust estimation of reporting rates, as well 

 as for improved estimation of catch-at-age numbers 

 in our model). However, while Polacheck and Hearn 

 (2003) found a relatively direct trade-off between the 

 level of observer coverage and number of tag releases 

 with their approximate model, we found with our 

 more comprehensive model that the trade-off depends 

 on the parameters of interest. In particular, greater 

 improvements could generally be achieved in the preci- 

 sion of the fishing mortality and cohort size estimates 

 by increasing the proportion of observer coverage than 

 by increasing the number of releases. On the contrary, 

 much larger gains were achieved in the precision of the 

 natural mortality-rate estimates by increasing the num- 

 ber of tag releases than by increasing the proportion of 

 observer coverage. Although the results will be highly 

 case-specific, these general observations were true in 

 all of the scenarios we considered, and we expect they 

 will hold true in a fairly wide range of scenarios. That 

 being said, the purpose of the simulations was not to 

 draw any specific conclusions, but to illustrate how the 

 model can be used to provide practical guidance about 

 the experimental design of a tagging study. 



A version of the BPO model has been used to 

 provide advice to the Commission for the Conservation 

 of Southern Bluefin Tuna (CCSBT) on the levels of 

 observer coverage and tag releases necessary to achieve 

 their objectives for a long-term tagging program 



conducted on SBT. To make the model more closely 

 resemble the situation for SBT it was necessary to 

 extend the model to a two-fishery situation with a 

 purse-seine fishery and a longline fishery, where tag 

 reporting rates were estimated from planted tags in 

 the purse-seine fishery and from observer data in the 

 longline fishery. Simulations, similar to those presented 

 here, were conducted with input parameter values that 

 best simulate the situation for SBT. The results showed 

 that the numbers of tags that were being released 

 each year were adequate, but that an increase in the 

 CCSBT's target level of observer coverage from 10% 

 to about 30% was required to meet the objectives of 

 the program regarding precision of the mortality-rate 

 estimates. 



In summary, the model presented here provides a 

 robust statistical framework for obtaining joint estimates 

 of mortality rates and abundance from tagging data in 

 situations where observers are present in the fishery. 

 The model can be used to provide insight into design 

 issues for those starting up new, or modifying current, 

 tagging and observer programs for the purposes of 

 estimating mortality rates and abundance. 



Acknowledgments 



We thank K. Pollock, D. Peel, and three anonymous 

 reviewers for constructive comments and suggestions 

 on drafts of this manuscript. The Australian Fisheries 

 Research and Development Corporation (FRDC) pro- 

 vided funding support for this research. 



Literature cited 



Akaike, H. 



1974. A new look at the statistical model identifica- 

 tion. IEEE (Inst. Electrical and Electronics Engi- 

 neers) Transactions on Automatic Control 19:716- 

 723. 



Brownie, C, D. R. Anderson, K. P. Burnhani, and D. S. Robson. 

 1985. Statistical inference from band recovery data: a 

 handbook, 2"<i ed., 305 p. U.S. Fish Wildl. Serv. Resour. 

 Publ. 156. 



Burnham, K. P. and D. R. Anderson. 



1998. Model selection and inference: a practical infor- 

 mation-theoretic approach, 353 p. Springer-Verlag, 

 New York, NY. 



Cormack, R. M. 



1964. Estimates of survival from the sighting of marked 

 animals. Biometrika 51:429-438. 

 Frusher, S. D., and J. M. Hoenig. 



2001. Estimating natural and fishing mortality and tag 

 reporting rate of southern rock lobster Jasi/s edwardsii 

 from a multiyear tagging model. Can. J. Fish. Aquat. 

 Sci. 58:2490-2501. 

 Hampton, J. 



2000. Natural mortality rates in tropical tunas: size really 

 does matter. Can. J. Fish. Aquat. Sci. 57:1002-1010. 

 Hearn, W. S., T. Polacheck, K. H. Pollock, and W. Whitelaw. 



1999. Estimation of tag reporting rates in age-struc- 



