Baremore et at: Gillnet selectivity for juvenile Carcharhinus hmbatus 
235 
Table 5 
Gillnet selectivity parameter estimates for each model for blacktip sharks ( Carcharhinus limbatus ) in northwest Florida, 1994- 
2010. All four models were run twice: first assuming fishing intensity to be equal across mesh sizes and again assuming that 
fishing intensity was proportional to mesh size. Model deviance is the likelihood ratio goodness of fit, with 130 degrees of freedom 
for each model. 
Equal fishing intensity Proportional fishing intensity 
Model 
Parameters 
Model deviance 
Parameters 
Model deviance 
Normal (fixed spread) 
(k, ct) = (5.98, 30.98) 
411.79 
(k, cr) = ( 6.94 , 34.91) 
371.36 
Normal (prop, spread) 
(a^ a 2 )=(6.8Q, 10.11) 
536.33 
(aj, a 2 ) = (8.11, 8.20) 
553.73 
Lognormal 
(Pj, ol = (4.00, 0.41) 
440.33 
(jUj, ct) = <4.17, 0.41) 
440.33 
Gamma 
(a, k) = ( 6.39, 1.17) 
469.68 
(a, k ) = ( 7 . 3 9 , 1.17) 
469.68 
was not necessary to test a separate method to estimate 
a gamma selectivity curve. 
Residual plots from all selectivity models showed 
some degree of bias for the smaller (50-70 cm FL) size 
classes in the 11. 4-, 12. 7-, and 14.0-cm mesh sizes. This 
finding indicated that all models underestimated the 
numbers of small blacktip sharks caught in these mesh 
sizes, and these underestimates could be an artifact of 
the sampling design of the GULFSPAN juvenile shark 
survey (Carlson and Brusher, 1999). In this survey 
gillnet panels were arranged in increasing order by 
mesh size, and the order of panels was not randomized. 
Randomization of gillnet panels is common in selectiv- 
ity experiments because it is thought to reduce the 
potential preference of fish for any one area of the net. 
However, because fixed stations were not used, and the 
nets were fished at a variety of depths, habitats, and 
seasons, sampling design was probably not a factor in 
the model’s lack of fit to the data. The overdispersion 
of the data could be a result of the pooling of the data 
into 5-cm bins, or could indicate schooling behavior by 
some size classes of blacktip sharks. Shark species are 
known to segregate by size and sex; therefore the cap- 
ture of a cluster of similar-size blacktip sharks is likely. 
Overdispersion does not necessarily affect parameter 
estimation (Millar and Fryer, 1999), although an initial 
model assumption may have been violated. 
Although the assumption of equal catches may have 
been violated, the second assumption of equal fishing 
effort among gillnet panels was most likely met. The 
shallow bays and estuaries sampled, along with the 
length of the net (-600 m), decreased the probability of 
different panels fishing in different habitats and depth 
zones. Commercial gear can be several kilometers in 
length, and sagging can cause the middle part of the 
gear to fish in different depth strata than those at the 
ends. Blacktip sharks were therefore equally likely to 
encounter each panel of the GULFSPAN survey gillnet. 
On occasion, adult blacktip sharks (>130 cm FL) have 
been captured in the survey areas on longlines (Bethea 
and Carlson 3 ). However, larger sharks are less likely 
to be caught in gillnets with mesh sizes smaller than 
20 cm, and those few large sharks captured in the 
smaller mesh sizes were generally entangled by roll- 
ing in the gear — a phenomenon that was also noted 
for finetooth sharks (C. isodon) (Carlson and Cortes, 
2003). All gillnet panels, except the 20.3 cm panel, were 
monofilament, and large sharks were able to break the 
monofilament and escape the gear. Such cases where 
larger sharks were entangled in smaller mesh sizes or 
where they broke free of the net could also have affected 
the lack of fit because the assumption of geometric 
similarity would not stand. The occurrence of larger 
sharks in small mesh sizes may have been reflected 
by the high model deviances for the models (normal 
proportional spread, lognormal, and gamma) where 
geometric similarity of the data was assumed. However, 
other than the lack of fit to the smallest size classes, 
the models described the data very well, with residu- 
als showing mostly equal error distribution and little 
systematic bias. 
Because of the change in the gear from 2005 through 
2006, several attempts were made to account for a 
year effect within the SELECT method. Because of 
low sample sizes within years, especially for the 7.6- 
and 20.3-cm panels, it was not possible to incorporate 
year as a factor. For instance, a total of 92 and 115 
blacktip sharks were captured by the 7.6- and 20.3-cm 
panels, respectively. Although these sample sizes were 
adequate for the overall model, when broken down by 
year the sample sizes were in the single digits for most 
size classes. The data were also separated into two 
time periods (1994-2005 and 2006-10), and the SE- 
LECT method was used to estimate selectivity models 
for each time period. The first time period produced 
reasonable results; however, no realistic solution was 
found for the second time period. This could also be 
due to sample sizes in the second time period. Although 
3 Dana M. Bethea and John. K. Carlson. 2011. Unpubl. 
data. Panama City Laboratory, Southeast Fisheries Sci- 
ence Center, National Marine Fisheries Service, National 
Oceanic and Atmospheric Administration, 3500 Delwood 
Beach Rd., Panama City, Florida 32408. 
