Terceiro and Ross: Estimation of age from length data for Pomatomus saltatrix 



541 



indication of how closely the alternative methods can 

 match the interpretation of the age structure commer- 

 cial landings of bluefish in North Carolina that would 

 be provided by the NCDMF 1986-1989 length-age test 

 data. By extension, the utility of these methods as 

 alternatives to a time series of geographically appro- 

 priate age-length keys can be evaluated on a quantita- 

 tive scale. 



Results 



Application of the alternative methods to the NCDMF 

 1986-1989 length data provided estimates of mean 

 length at age that were lower than the NCDMF value 

 for age-0 fish, and generally higher for ages 1 and 

 older. Mean lengths at age estimated by MULTIFAN 

 for ages 3 to 5 were the exception, as they were lower 

 than the NCDMF values (Table 3). 



For the annual and combined NCDMF 1986-1989 

 length distributions, cohort slicing using the NOAA 

 (1989) growth parameters consistently provided the 

 poorest match with the NCDMF proportions at age 

 (Tables 4-8, pages 543-547); there were large differ- 

 ences relative to the NCMDF data for ages 0, 4, and 5. 

 Differences were significant (P > 0.05) in 1988 and for 

 the combined distribution. 



Standard application of the CNN ALK provided im- 

 proved results over simple cohort slicing, and estimated 

 age proportions were not significantly different from 

 the true NCDMF proportions, except when data were 

 combined. Estimated proportions at age for fish of less 

 than 40 cm were very close to the NCDMF propor- 

 tions. The largest value of D always occurred at age 4 

 or 5, reflecting problems in resolving fish in the 60- 

 75 cm length interval to the correct age when using 

 the standard age-length key application of the CNN 

 length-age data (Tables 4-8). 



The IALK method provided initial results that were 

 surprisingly poor. The convergence criteria proposed 

 by Kimura and Chikuni (1987) for IALK estimated 

 proportions at age were intended to stop iteration of 

 the key when the underlying age distributions of the 

 length-age data and the length frequency to be aged 

 closely matched, thus better satisfying the conditions 

 necessary for unbiased application of the key. 



Two characteristics of bluefish biology, multiple 

 spawned cohorts for the same year class and differen- 

 tial availability to fisheries by age class, appear to 

 have combined to result in enough difference in the 

 length distribution patterns of the CNN and NCDMF 

 data to significantly hinder the effectiveness of the 

 IALK model when the final iteration of the key was 

 used. Iteration of the annual CNN length-age distri- 

 butions reduced the number of bluefish in the 50-60 cm 



interval, and increased the number of bluefish in the 

 20-40 cm (ages and 1) and 65-75 cm (ages 4 and 5) 

 length intervals. The resulting length and age 

 distributions of the input CNN length-age data 

 matched the NCDMF lengths more closely than the 

 initial distributions. However, when the final iterated 

 key was applied to NCDMF lengths, the cumulative 

 proportions at age generally provided large values 

 of D for ages and 1, or ages 4 and 5 (Table 9, page 

 548). 



Intermediate iterations of the CNN length-age data 

 (generally the first or second iteration) provided a bet- 

 ter match to the NCDMF proportions at age than the 

 final iteration (generally about the 15th iteration). In 

 effect, we limited the degree to which the CNN length- 

 age distribution could be changed by evaluating the 

 application of each iteration of the CNN ALK, and 

 selecting as the best solution the estimated propor- 

 tions at age providing the smallest cumulative differ- 

 ence from the NCDMF cumulative proportions. Under 

 this procedure of evaluation, the IALK provided the 

 same or slightly improved results relative to standard 

 application of the CNN length-age data. Only for the 

 combined 1986-1989 length distributions were the 

 IALK estimated proportions at age significantly differ- 

 ent from the NCDMF proportions at the 5% level 

 (Tables 4-8). 



MULTIFAN results were inconsistent when con- 

 sidered on an annual basis. The growth parameters 

 selected by MULTIFAN as best describing the 

 growth pattern of bluefish during 1986-1989 (Table 

 10, page 549) performed relatively poorly in matching 

 the NCDMF 1986 and 1987 proportions at age, but 

 provided a close match to the NCDMF 1988 and 1989 

 proportions (Tables 4-8). Variation in growth and 

 recruitment to the sampled fishery among cohorts 

 contributed to varying success in matching the 

 annual NCDMF proportions with the MULTIFAN 

 method. 



The best fitting MULTIFAN run provided the only 

 non-significant value of D (at age 6, the "plus group" of 

 the MULTIFAN model interpretation) for the combined 

 1986-1989 length distributions obtained among the 

 compared methods. The largest values of D provided 

 by the other methods occurred at ages 4 or 5, indicat- 

 ing problems in resolving the ages of fish 60 cm and 

 larger in accordance with the NCDMF interpretation. 

 Only MULTIFAN provided proportions at ages 4 and 5 

 that accurately reflected the NCMDF proportions in 

 the combined distributions (Table 8). The improved 

 performance of MULTIFAN when the NCDMF data 

 are evaluated in combined fashion reflects better ad- 

 herence to the underlying concept of the MULTIFAN 

 model (Fournier and Breen, 1983). 



