Musyl et al.: Postrelease survival, vertical and horizontal movements, and thermal habitats of five species of pelagic sharks 
343 
pop-up dates were set at 8-13 months after deployment 
of the tags. Depth and temperature data were mea- 
sured as 8-bit numbers, yielding a depth resolution of 
~5.4 m and temperature resolution of ~0.17°C. Fail-safe 
options were also programmed into the PSAT software 
whereby stationary PSATs (i.e., those experiencing no 
significant changes in pressure) or shed tags would 
begin transmitting archived data to the ARGOS satel- 
lite system after four days. In the event of mortality, 
once the shark sank to -1200 m and remained there for 
-15 minutes, the PSAT would separate from the shark, 
float to the surface, and begin transmitting stored data 
to ARGOS. 
Daily (raw) geolocation estimates were calculated by 
the manufacturer using ambient light-level irradiance 
data during postprocessing of the satellite data with a 
proprietary algorithm (Gunn and Block, 2001). From 
the raw geolocations, most probable tracks (MPTs), 
movement parameters, and associated error estimates 
were calculated by a state-space Kalman filter algo- 
rithm with position estimates refined with the use of 
sea surface temperature (SST) (Nielsen et al., 2006). 
Depth and temperature data were assigned to daytime 
or nighttime according to times of local dusk and dawn 
derived from longitude and latitude estimates (from the 
MPTs) with the use of standard astronomical formulae 
(Meeus, 1998). 
Resampling techniques were used to construct 95% 
parametric bootstrap confidence intervals (Cl*) (with 
the assumption of a binomial distribution with 10,000 
replicates) for postrelease mortality estimates and 
PSAT reporting rates (Manly, 2007). Meta-analysis 
was used to estimate a summary effect for postrelease 
mortality in blue sharks from published studies (Weng 
et al., 2005; Campana et al., 2009a; Stevens et al., 
2010) and the present report, by assuming that these 
studies represent random samples of some population in 
which the underlying (infinite-sample) effect sizes have 
a distribution rather than a single value (i.e., random 
effects model, Borenstein et al., 2009). The analysis was 
conducted on the logit (log odds ratio) of the proportion 
of blue sharks that ultimately died as identified from 
PSATs across studies by using Comprehensive Meta 
Analysis, vers. 2.2 (www.Meta-Analysis.com, accessed 
November 2010). Postrelease mortality estimates and 
95% confidence intervals were weighted by sample size 
and the number of studies where heterogeneity was as- 
sumed (i.e., with the random-effects model where each 
study was assumed to have its own postrelease mortali- 
ty rate and variance). The Q statistic, a measure of het- 
erogeneity, was calculated to test whether postrelease 
mortality estimates across studies were similar, and the 
Z test was used to determine whether the postrelease 
mortality estimate was significantly greater than zero 
(Borenstein et al., 2009). If postrelease mortality is 
consistent across studies, then the meta-analysis yields 
a combined estimate that is more precise than any of 
the separate estimates (Borenstein et al., 2009). For 
presentation purposes, logits were converted back into 
percentages. 
Data provided by the PSATs were divided into six 
data streams by parsing depth data into day depth 
(DD), night depth (ND), and “all” depth (=both day and 
night) (AD); and temperature data into day tempera- 
ture (DT), night temperature (NT), and combined tem- 
perature (AT). Nonparametric tests were used to exam- 
ine variation by species with Kruskal-Wallis ANOVAs 
(to compare equality of medians across individuals) for 
each of the data streams where the test statistic ( H c ) 
was adjusted for ties (Zar,1996) because data distri- 
butions were not normally distributed (Lillifors tests, 
PcO.Ol). For each species, multiple post-hoc pairwise 
Mann-Whitney W-tests, with Bonferroni corrected P- 
values to account for inflation of type-I error based 
on multiple tests of the same hypothesis (MWBC), 
were used to compare equality of medians within and 
between individuals for each of the data streams (Zar, 
1996). When only a single Mann-Whitney test could be 
performed, Monte Carlo methods (10,000 random as- 
signments) were used to obtain an empirical P-value 
that approximated the exact P-value without reliance 
on asymptotic distributional theory or exhaustive enu- 
meration (Manly, 2007). The greatest vertical dis- 
tance (D max ) between cumulative distribution functions 
among tags from two-sample Kolmogorov-Smirnov 
(KS) tests was formatted into distance matrices as 
input for the unweighted pair-group method by using 
arithmetic average (UPGMA) clustering (Sneath and 
Sokal, 1973; Musyl et al., 2003). This procedure al- 
lowed us to observe patterns of depth and temperature 
preferences across pelagic species. Electronic tag data 
from Pacific bigeye tuna ( Thunnus obesus ) (Musyl et 
al., 2003), swordfish, black marlin ( Istiompax indica), 
and blue marlin ( Makaira nigricans ) (Musyl et al. 2 ) 
served as outgroups to help clarify and define rela- 
tionships (Sneath and Sokal, 1973). The cophenetic 
correlation was used as a measure of goodness-of-fit 
between the matrices and resultant clustering den- 
drograms (e.g., 0.7-0. 8 is considered “poor,” >0.8 is 
considered “good,” and >0.9 is considered “very good” 
[Rohlf, 1992]). 
Time-at-depth and time-at-temperature data were 
aggregated into 20-m and 1°C bins, respectively. These 
data were subsequently expressed as a fraction of the 
total time of observation for each shark, and the frac- 
tional data bins were averaged across all sharks within 
each category. For sharks experiencing several lunar 
cycles, the correlation coefficient ( R ) was determined 
between average nighttime depth (m) and lunar illu- 
mination (Zar, 1996). Lunar illumination data were 
obtained from the United States Naval Observatory 
(http://aa.usno.navy.mil/data/docs/MoonFraction.php, 
accessed June 2010) and were uncorrected for cloud 
2 Musyl, M. K., L. M. McNaughton, J. Y. Swimmer, and R. W. 
Brill. 2004. Convergent evolution of vertical movement 
behavior in swordfish, bigeye tuna and bigeye threshers. 
Vertical niche partitioning in the pelagic environment as 
shown by electronic tagging studies. Pelagic Fisheries 
Research Program, Univ. Hawaii Manoa, Newsletter 9:1-4. 
