238 
Fishery Bulletin 117(3) 
17 specimens were heterozygous at the Sma6 locus and 
were identified as hybrids, and 70 samples were degraded 
or otherwise produced inconclusive results. The data col¬ 
lected in 2013 indicate an improvement to 94% and 68% 
accuracy in field identifications of blackspotted rockfish 
and rougheye rockfish, respectively. Of the 902 specimens, 
428 were blackspotted rockfish, 429 were rougheye rock¬ 
fish, 8 were Sma6 hybrids, and 37 fish could not be iden¬ 
tified to species. 
Multivariate regression and length at age 
Across all measurements examined, otoliths of rougheye 
rockfish tended to be a larger size for a given age than black¬ 
spotted rockfish, as shown by bivariate plots (Fig. 2). This 
pattern manifests as a small but significant differences in 
the initial size of otoliths and in the rate at which these struc¬ 
tures grow with age. According to all 7 of 
the parameters, otoliths of rougheye rock¬ 
fish are greater beginning in the first few 
years of life, as evidenced by a significant 
species effect on the intercept (Table 3). 
A more varied response is observed as 
the fish grow, with some characteristics 
like otolith length, major axis length, 
and perimeter tending to converge over 
time (Fig. 2). By comparison, otolith area 
becomes even more divergent. In most 
cases, the difference in slopes between 
the species is not significant (Table 3), 
indicating that most of the size difference 
in rougheye rockfish occurs in the early 
years of life. A very similar dynamic is 
observed in the von Bertalanffy growth 
curves (Fig. 3, Table 4). Rougheye rockfish 
have a much higher rate of growth early 
in life, leading to a significant gap in the 
length at age between the 2 species. 
Classification success 
Logistic regression found that several 
otolith parameters and age interactions 
were predictive of species between these 
2 rockfish species. Several of the predic¬ 
tors used in this model are highly cor¬ 
related; however, removal of these 
collinear predictors resulted in a worse 
fit to the data as measured by AIC. 
Removing collinear predictors had little 
effect on discrimination accuracy. There¬ 
fore, only nonsignificant parameters 
that did not increase AIC were elimi¬ 
nated from the final model. The full 
model that includes all 8 parameters 
and 7 interactions had an AIC of 143.35, 
and the reduced model that includes all 
8 parameters and 2 interactions had an 
AIC of 139.93 (Table 5). Of the 638 specimens used to 
build the logistic regression, 94.7% of 434 blackspotted 
rockfish and 88.2% of 204 rougheye rockfish could be cor¬ 
rectly identified, once uncertain cases were removed 
(Table 6). However, the best test of a classification 
method is its accuracy when confronted with new data. 
When used to classify 221 specimens from the testing 
data set, it was successful for 97.3% of 112 blackspotted 
rockfish and 86.2% of 109 rougheye rockfish (Fig. 4, 
Table 6). Overall, 11 of the 221 fish are classified as 
uncertain because their prediction intervals indicate 
both possibilities of species identification are reasonably 
likely. Of these 11 specimens, 5 fish would have been 
assigned to the incorrect species, a result that would 
almost double the total number of misidentifications. 
When uncertain cases were excluded, only 0.9% of black¬ 
spotted rockfish and 5.5% of rougheye rockfish were mis- 
identified (Table 6). 
Age 
Figure 2 
Bivariate plots with species-specific regression lines for the relationships of 
different otolith morphometries with age from models based on data from 
rougheye rockfish (Sebastes aleutianus ) (open circles, dashed lines) and black¬ 
spotted rockfish iS. melanostictus ) (black circles, solid lines) collected in the 
Gulf of Alaska in 2009 and 2013. Note that age is shown on a logarithmic scale. 
Across all measurements, rougheye rockfish are larger at any given age than 
blackspotted rockfish. 
