156 
Fishery Bulletin 106(2) 
were in gross otolith morphological features or low- 
order Fourier harmonics. 
Sex and year-specific linear discriminant functions 
yielded a range of shape variables selected, and the 
mean accuracy of classifications ranged from 65.8% to 
76.4% among models (Table 1). Discriminant functions 
included between five and seven variables. The highest 
classification accuracies from a given model were 71.1% 
for GOM females and 81.7% for Atlantic females in 
2001 (mean accuracy 76.4%). The lowest classification 
accuracies were 61.2% for GOM males and 70.4% for 
Atlantic males in 2002 (mean accuracy 65.8%). Clas- 
sification accuracies were slightly higher for Atlantic 
fish (67.9-81.7%) than for GOM fish (61.2-71.1%) for 
most models. 
Atlantic stock king mackerel contributed to landings 
in all three winter sampling zones. Maximum likelihood 
models estimated that the contribution of Atlantic fish 
to winter landings ranged from 14.5% for females in 
zone 1 in 2002 to 99.9% for males in zone 2 in 2001 
(Table 2). Bias-corrected bootstrapped 90% confidence 
intervals varied among zones and between years but 
generally were on the order of point estimates ±20%. 
Bootstrap cumulative frequency distributions demon- 
strated that although the majority of bootstraps fell 
near point estimates, wide confidence intervals resulted 
from long upper and lower distribution tails (Figs. 5 
and 6). 
The estimated contribution of the Atlantic stock to 
2001-02 winter landings was similar between males 
and females among all three winter sampling zones, 
except for zone 2 where few males were sampled (Table 
2). In winter 2002-03, Atlantic females contributed 
less than males and also had lower contribution es- 
timates than females in 2001-02. Atlantic males had 
similar contribution estimates during both sampling 
years. Overall, a gradient in contribution estimates 
was observed; there were higher Atlantic stock percent- 
ages in southeast Florida (zone 
3) and declining Atlantic stock 
presence in southwest Florida 
landings (zone 1). 
Discussion 
Classification accuracies from 
stepwise linear discriminant 
function analysis confirm the 
feasability of using otolith-shape 
parameters to distinguish king 
mackerel stocks but also dem- 
onstrate that stock-specific oto- 
lith-shape parameters provide 
natural tags that are far from 
perfect (i.e., <=100% stock dis- 
crimination success). The clas- 
sification success that we report 
(61.2% to 81.7%) is similar to the 
range reported in shape-based 
stock or population discrimina- 
tion for other fishes (e.g., 54.9% 
to 89.3% for lake whitefish, Core- 
gonus clupeaformis [Casselman 
et al., 1981]; 63.9% to 87.5% 
for Atlantic salmon [Friedland 
and Reddin, 1994]; 61% to 83% 
for haddock [Begg et al., 2001]; 
and, 63.6% to 83.3% for coral 
trout, Plectropomus leopardus 
[Bergenius et al., 2006], as well 
as that previously reported by 
DeVries et al. [2002] for female 
king mackerel [65.8% to 85.7%]). 
However, the lack of more dis- 
tinct differences in otolith shape 
between stocks likely contrib- 
uted significantly to the wide 
confidence intervals estimated 
60 
50 
40 
30 
20 
10 
0 
B 
Females 
Males 
Figure 4 
Age distribution of king mackerel (Scomberomorus c avalla) in samples collectged 
in winters 2001-02 and 2002-03. (A) = 2001 zone 1 ; (B) = 2002 Zone 1 ; (C) = 
2001 zone 2 ; (D) = 2002 zone 2 ; (E) = 2001 zone 3 ; and (F) = 2002 zone 3. 
