242 
Fishery Bulletin 11 7(3) 
' 
A 
Wm 
cmsn ~-2 
0.00 0.1 
Error (CV) 
Figure 5 
Heatmap showing the accuracy of the species prediction method used in this 
study, with simulated errors. Lighter shades of gray indicate higher accuracy. 
Contour lines highlight the gradual change in values of accuracy. Coefficients 
of variation (CVs) on the x-axis represent simulated random errors in the 
aging process, and values on the y-axis indicate bias in a specific positive or 
negative direction. 
grow faster than otoliths of blackspotted rockfish. If ages 
are systematically biased upward, more fish will appear 
to be slower growing and the model will predict a greater 
number of blackspotted rockfish. Conversely, a bias 
toward lower ages will make specimens appear to be 
faster growing and will tend to predict more rougheye 
rockfish. With this in mind, it is imperative that the age 
determination process carefully police any nonrandom 
errors. 
This model is not designed to identify hybrids or clas¬ 
sify hybrids. Of the 1847 specimens that were genetically 
identified from the surveys conducted in 2009 and 2013, 
25 (1.3%) fish were hybrids at the Sma6 locus. With such 
a small sample, this study was unable to determine what 
potential effect these hybrids might have had on the 
accuracy of identifications. The best course of action may 
be to assume that hybrids represent a minor source of 
identification error. Furthermore, this model should not 
be expected to identify or remove miscellaneous rockfish 
species from a mixed sample of otoliths. All specimens 
must be correctly identified as part of the rougheye and 
blackspotted species complex in the field, and it will 
require that field personnel be adequately trained to 
identify other species. 
All the specimens used in this study were collected 
from the Gulf of Alaska, but the rougheye and black¬ 
spotted rockfish species are found together off the 
western coasts of the United States and Canada, in the 
eastern Aleutian Islands, and in the 
southern Bering Sea. We have tried to 
account for any potential geographic 
and temporal differences by collecting 
specimens from different locations in 
the Gulf of Alaska (Fig. 1) and in 2 dif¬ 
ferent years. The model was developed 
and fit by using data from 2009, but 
it was tested against additional data 
from 2009 as well as data from a second 
collection of specimens made in 2013. 
The model performed well with data 
from both years, with only 2-3% dif¬ 
ference between years in overall accu¬ 
racy. We think this result is a strong 
indication that attempts to apply this 
method in future years throughout the 
Gulf of Alaska and Bering Sea will be 
successful. 
Considering the positive perfor¬ 
mance of this logistic regression model, 
we predict it will be useful to classify 
or reclassify historical otolith samples 
from mixed species catches. Applying 
these findings backward in time does 
rely on the assumption that the growth 
rates of these fish species in the present 
day are similar to those of the past and 
have not been altered by changes in cli¬ 
mate, fishing, or other forces. However, 
given that both species have a rather 
long life span, many of the fish in this study did experi¬ 
ence those past environmental conditions. Many of the 
specimens collected are 30 years old or older, and the 
oldest is a 135-year-old rougheye rockfish with an esti¬ 
mated birth year of 1878. Errors for the model were not 
correlated with increasing age, meaning that the model 
was capable of assigning the correct species whether 
the majority of a fish’s growth occurred in present day 
conditions or those of decades past. This is a promising 
indication that the model will be reliable for archived 
specimens going back at least 20-30 years, making it 
invaluable for establishing individual stock assessments 
for each species. 
The rougheye rockfish and the blackspotted rockfish 
are assessed as a complex in a single statistical age- 
structured model that assumes equal age compositions, 
growth rates, and many other parameters (Shotwell et al., 
2017). However, it has been observed that the spatial and 
depth distributions of the 2 species are different (Orr and 
Hawkins, 2008) and that rougheye rockfish appear to 
grow faster and mature earlier (Conrath, 2017). Although 
Conrath’s findings are consistent with our observations 
of growth at age, her study did not confirm species iden¬ 
tifications with genetics, raising the possibility that her 
estimates are skewed by misidentifications. In the future, 
our method will allow researchers to more confidently 
identify these species and arrive at more accurate esti¬ 
mates of important population parameters. New studies, 
