Estimating survival and exploitation rates from a tagging study on juvenile Atlantic halibut

3.6 Model Assessment

1. 
   Releases occur at an instant in time. The majority of releases occur in the 4 month period May-August of each year. We assume that all releases took place on 1 May of the respective recovery year and is available to the fishery for the next 12 months. However, if fish released in August are not available to the fishery before their release leading to a smaller than expected number of tags recovered. This may be reason for the lack of fit in the later release years.
   

2. 
   Fishing occurs uniformly throughout the year. We adopted the equal fishing effort formulation of Hoenig et al (1998). Table 6 shows that while some recoveries took place over the entire year, the fishing effort appears to be reduced in the later part of the calendar year and is higher in the summer months. Hoenig et al (1998) looked at the effects of different models for the fishing pattern, and found that there were only modest effects on the estimates.
   

3. 
   Mixing with untagged fish. In order that exploitation rates of tagged fish reflect general exploitation rates, it is necessary that tagged fish mix with untagged fish prior to exploitation. In some cases, newly tagged fish do not have sufficient time to mix with untagged fish in the first year of the fishery. Biases in expoitation rates can be positive (negative) depending if newly tagged fish are more(less) likely to be captured. Biases on survival rates are harder to quantify as survival rates use multiple year and cohorts and presumably the non-mixing is small. Hoenig et al (1998b) developed models that allowed for incomplete mixing. They concluded that The simulation results confirm that there is a substantial penalty for having to use a non-mixed model with estimates having poor precision. We did not fit such a model to this dataset because other violations of assumptions are likely to be more serious.
   

4. 
   No initial tagging induced mortality and 100% tag retention (). We understand that some fish were retained in pens to assess the initial tagging mortality and shedding assumption and the estimated loss rates are small. Unfortunately, long term tag retention can only be estimated by double tagging experiments which was not done in this experiment. Tag loss is indistinguishable from natural mortality and estimated natural mortality rates will tend to be underestimated in the presence of tag loss.
   

5. 
   100% tag reporting. We assumed 100% tag reporting in order to estimate exploitation rates. Low tag-reporting rates tend to deflate tag exploitation rates (fewer tags are reported) but have little effects on estimates of natural survival. The tag reporting rate can be estimated using reward tags which were not applied until very late in the study. The (very sparse) data indicates a very low tag-reporting rate so the reported exploitation rates may not be realistic. We attempted to fit the Hoenig et al (1998) model assuming a tag-reporting rate of 0.10, but the model lead to nonsensical estimates of mortality and exploitation and are not reported.
   

6. 
   Homogeneous survival and exploitation – The tag-recovery models of Brownie et al (1985) and Hoenig et al (1998) assume that survival and exploitation rates are homogeneous across the tagged animals. This study is a mixture of different sizes and different sexes of halibut caught in wide set of locations. Plots of the locations of releases and recoveries (not shown) appear to indicate good mixing along the Scotian shelf. Plots of reported lengths show released halibut are primarily in the 70-80 cm length class. Because fish grow with time, there is a very strong relationship between the increase in length and the time at liberty, with recaptured fish ranging in size from 80-150 cm (primarily in the 80-100 cm length class). Pollock and Raveling (1982) and Nichols et al (1982) investigated the robustness of Brownie model to heterogeneity in survival and recovery and found that these models were robust (i.e. estimates reflected the average survival and tag-recovery rates) except where there was a strong relationship between survival and recovery rates (e.g. a fish get longer they have higher survival and recovery rates).
   

7. 
   All tags correctly identified when applied and recovered. Substantial release records do not have a date of release. Some of these records appear to below to lost tags (e.g. lost on swamping of a vessel), but the fate of these other tags is unknown. Without knowing the date of application, these release records cannot be used. If this missing information occurs at random, then these records can be ignored and no biases occur (it looks as if they were simply not applied), but precision is reduced. There is some indication that some of the tags without release dates were applied to fish as some were “recovered”. Similarly recovered tags cannot be used without the date of recovery. If these occur at random, then there is no effect on survival rates, but estimates of exploitation rates are biased low because fewer tagged fish are reported. Some tag numbers could not be read and these were also ignored. If this occurred at random, estimates of survival are unbiased but estimated exploitation rates are again biased low.
   

8. 
   Tagging young halibut only. Brownie et al (1985) caution strongly against banding young animals only. The primary reason is that younger animals often have both a differential survival and tag-recovery rate. These two parameters become completely confounded and cannot be estimated and also caught adult survival rates and exploitation rates to become non-estimable. This model assumes that all animals, regardless of age (or length) have the same survival and exploitation rates, but we suspect the data is too sparse to investigate effects of length (or other covariates) to verify this assumption.