18 
Fishery Bulletin 11 7(1-2) 
ture event (e.g., with mortality caused by predation). 
Consequently, some researchers have restricted their 
analysis to the first 5-10 d after tagging (Graves et al., 
2002; Kerstetter et al., 2003; Horodysky and Graves, 
2005; Marcek and Graves, 2014); however, given the 
relatively low natural mortality rate of adult bluefin 
tuna relative to other species (Fromentin and Powers, 
2005), we extended the time frame from 5-10 d to 30 
d, following Stokesbury et al. (2011). The penalty for 
use of a longer time frame is that natural (and fishing) 
mortality begins to bias estimates of release mortal¬ 
ity, but it would be highly unlikely (0.8% chance) for 
a bluefin tuna to not survive 30 d because of natural 
mortality, given the currently assumed natural mortal¬ 
ity rate for bluefin tuna (0.1/year; ICCAT, 2017). 
Any fish that appeared to live past this 30-d thresh¬ 
old was determined to have successfully survived the 
capture event. Wilson score intervals and 95% confi¬ 
dence intervals (CIs) were calculated for the binomial 
proportions (Wilson, 1927). The standard method for 
determining mortality by using PSATs involves infer¬ 
ring mortality from behavior of the fish as recorded 
by the tag. Below, we outline this method; however, 
in this study, we had to address an added complica¬ 
tion. To distinguish between a mortality event and a 
premature tag release, we considered whether a tag 
floated after it was shed by a fish or it remained on 
a dead fish until it reached either the release depth 
of the RD-1800 device or the programmed tag release 
time. Given the ability of bluefin tuna to swim at high 
rates of speed (Wardle et al., 1989) and because fish 
remained in the water for tagging that occurred at 
night often on poorly lit vessels and varying states of 
sea conditions, some premature tag shedding was like¬ 
ly to have occurred in our study, and it is commonly 
observed in most PSAT tagging studies (Musyl et al., 
2011b). The negative buoyancy associated with some of 
the deployed tags complicates the interpretation of the 
recorded depth data because a tag attachment failure 
would result in the tag sinking in a similar fashion to 
a dead fish. In all, 27 of the 41 deployed PSATs were 
negatively buoyant. 
Despite the negative buoyancy of those tags, we 
were able to distinguish between likely premature re¬ 
lease of a tag and a fish mortality by calculating the 
sinking rates for each of the 10 prematurely released 
tags and comparing these rates to the rate (0.251 m/s) 
for the tag that was dropped overboard (the reference 
tag). In addition, there was an apparent mortality of a 
bluefin tuna that was tagged with a positively buoyant 
PSAT and tether; the sinking rate calculated for this 
tag was 0.408 m/s, a rate that is over 60% faster than 
the rate of the reference tag. Assuming that all dead 
fish would sink at a faster rate than the reference tag, 
we classified each tag according to whether it likely 
sunk because of a premature release (sinking rate<rate 
of reference tag) or because of a fish mortality (sinking 
rate>rate of reference tag). Fish were then assigned to 
1 of 4 categories on the basis of the observed behavior 
of the fish as recorded by the tag: 1) survived (consis¬ 
tent vertical movement for >30 d), 2) mortality (fish 
was at large for <30 d, tag detachment occurred at a 
depth >1200 m, and the sinking rate was >0.251 m/s), 
3) tag attachment failure (fish was at large for <30 d, 
and the tag was positively buoyant and detached at a 
depth <1200 m, or the tag was negatively buoyant and 
detachment occurred at a depth >1200 m, but the sink¬ 
ing rate was <0.251 m/s), and 4) non-reporting tag (tag 
failed to transmit any data). 
To account for the uncertainty of the eventual fate 
of fish that were equipped with tags that either failed 
to report or failed to remain attached for >30 d, we 
calculated the mortality rate by using 2 methods. One 
method used this expression that includes the num¬ 
ber of fish assigned to 3 of the 4 categories, yielding 
the highest possible mortality estimate: (mortality-i-tag 
attachment failure+non-reporting tag)/total number of 
tags deployed. The other method used the following 
expression: mortality/)survived+mortality). The first 
method assumes that all fish in the tag attachment 
failure and non-reporting tag categories were dead fish, 
but the second effectively considers that tag data for 
fish in the non-reporting tag or tag attachment failure 
categories are uninformative and discards those fish 
from the sample. 
Estimation of overall mortality 
Overall mortality (M) was calculated as the probability 
of a mortality occurring during the entire capture and 
release process. It is calculated as the probability of be¬ 
ing dead at-vessel ( P(C )) times the probability of dying 
after being released ( P(R )): 
M = P(C)xP(R). (1) 
The variance of estimates from this equation was de¬ 
rived as the variance of the product of 2 assumed un¬ 
correlated random variables (Goodman, 1960). 
Results 
Pelagic Observer Program database 
The results of the logistic regression found that only 
one variable of the independent model, hook type, sig¬ 
nificantly (P<0.05) affected the probability of at-vessel 
mortality for bluefin tuna (Table 1). Therefore, we re¬ 
port the least square means as estimates of at-vessel 
mortality rate for the 3 hook types, standard (strong) 
circle hook (65%, 95% Cl: 57-72%), J-hook (68%, 95% 
Cl: 56-78%), and the currently mandated weak hook 
(54%, 95% Cl: 46-62%). 
Tagging 
From 2010 to 2015, 41 adult bluefin tuna from PLL 
vessels were tagged with Wildlife Computers PSATs in 
the GOM (Table 2). The size range of the 41 bluefin 
tuna was 190-270 cm straight fork length (Table 2), 
