Harris et al.: Otolith morphometric analysis for species discrimination of Sebastes melanostictus and 5. aleutianus 
237 
Table 1 
Division of data from blackspotted rockfish ( Sebastes mel¬ 
anostictus) and rougheye rockfish (S. aleutianus) collected 
in the Gulf of Alaska in 2009 and 2013 into the 2 data 
sets used to construct and validate the logistic regression 
model. The model parameters were fit by using specimens 
from the training data set and then tested against a new 
set of specimens in the testing data set. The training data 
set includes 638 specimens, all from 2009, and the testing 
data set includes 221 specimens from both 2009 and 2013. 
Number of specimens 
Species 
Year 
Training 
data set 
Testing 
data set 
Total 
Blackspotted 
2009 
434 
50 
484 
rockfish 
2013 
0 
62 
62 
Rougheye 
2009 
204 
50 
254 
rockfish 
2013 
0 
59 
59 
Total 
638 
221 
859 
could not be genetically identified to species or because 
the corresponding otoliths were damaged and could not be 
measured. Of the fish used in the study, 50 specimens from 
each species were randomly assigned to the testing data 
set, and the remaining 638 fish were used to construct 
and fit the logistic regression function. Not all of the 902 
specimens from the survey conducted in 2013 were nec¬ 
essary for an adequate test; therefore, 121 fish were ran¬ 
domly chosen and they were included in the testing data 
set. Overall, the training and testing sets covered similar 
ranges of age and length values, and the training and test¬ 
ing sets had similar mean ages and lengths (Table 2). In 
the testing data set, the predicted species of each otolith 
was compared with the genetically determined identity to 
yield the accuracy of this classification method. 
This method relies heavily on otolith age to help dis¬ 
criminate between the blackspotted and rougheye rockfish 
species, but otolith age is typically more prone to measure¬ 
ment error than standard methods. Furthermore, these 
species frequently attain very old ages and are among 
the most difficult to read. Therefore, we simulated the 
potential effect of errors in the age reading process by gen¬ 
erating a false age and using that age to estimate new clas¬ 
sification probabilities. This simulation looks at 2 types of 
errors: bias and random noise. Bias is a systematic ten¬ 
dency to age an otolith higher or lower than its true age. 
With a -10% bias, a 10-year-old fish would be evaluated 
as if it were a 9 years old, and a 100-year-old fish would 
be evaluated as if it were 90 years old. Random noise was 
simulated by drawing from a normal distribution defined 
by the age and a coefficient of variation (CV) (standard 
deviation [SD]/mean). Hence, a 10-year-old fish with a CV 
of 0.1 will have its age adjusted in each simulation run by 
an amount drawn from a normal distribution with an SD 
of 1, and an SD of 10 would be used for a 100-year-old fish. 
These types of errors are combined to produce a simulated 
age by using this equation: 
New Age ~ N(\x = MeasuredAged + bias), 
o = MeasuredAge x CV ), 
where N is the normal distribution with the mean equal 
to p and the standard deviation equal to o. This process 
was repeated 5000 times for each combination of error and 
bias, with the average classification accuracy measured 
for each run. All statistical analyses for this simulation 
were carried out by using R. 
Results 
Genetics 
The results of the genetic analysis from the data collected 
during the survey in 2009 show that field identification 
accuracy was 92% for blackspotted rockfish and 66% for 
rougheye rockfish. Out of 945 samples, 540 fish were 
genetically identified as blackspotted rockfish and 318 
fish were identified as rougheye rockfish. Additionally, 
Table 2 
Summary statistics for specimens of blackspotted rockfish ( Sebastes melanostictus) and rougheye 
rockfish (S. aleutianus) sampled for the training data set used to construct the logistic regression 
model and for the testing data set used to validate the model’s accuracy. Specimens were collected 
in the Gulf of Alaska in 2009 and 2013. All lengths are fork lengths in millimeters, and all ages are 
in years. n=number of specimens. 
Species 
Data set 
n 
Length 
Age 
Min. 
Mean 
Max. 
Min. 
Mean 
Max. 
Blackspotted rockfish 
Training 
434 
70 
355 
570 
2 
22 
113 
Testing 
112 
130 
348 
540 
3 
22 
103 
Rougheye rockfish 
Training 
204 
130 
396 
670 
3 
20 
135 
Testing 
109 
120 
371 
730 
3 
17 
83 
