Labelle et al Determination of age and growth of South Pacific albacore 



655 



>95 c 7c of the cross-checks; samples were rejected if the 

 discrepancies in counts could not be resolved. A linear 

 relationship (r-=0.94) was detected between ring counts 

 for the 35th and 36th vertebrae of the first 200 alba- 

 core processed, and no significant difference in age 

 composition was found between counts from the two 

 sets of vertebrae (\ 2 test, P=0.9). Even though the 35th 

 and 36th vertebrae were both suitable for study, we 

 chose the 35th vertebrae because it was larger and 

 easier to read in most cases. 



Tagging operations 



For tagging, albacore were caught mainly with com- 

 mercial troll gear (Dotson, 1980). The tagging proce- 

 dure was similar to that described by Laurs et al. 

 ( 1976). Immediately after a strike, the fishing line was 

 retrieved manually. Once the specimen was on board, 

 the hook was removed and the albacore was quickly 

 inspected for injuries to the gills, eyes, mouth, and 

 palate. Albacore that were vigorous and not visibly 

 injured were placed on a measuring board or tagging 

 cradle (Kearney and Gillett, 1982) and were tagged as 

 rapidly as possible with a stainless steel tube applica- 

 tor that contained a 13-cm-long serially numbered spa- 

 ghetti-type plastic tag with a single barbed nylon head. 

 The tag was inserted at an oblique angle so that the 

 barb was anchored among the pterygiophores of the 

 second dorsal fin. Fork length was measured and re- 

 corded on audio tape, along with the date, time, alba- 

 core condition, and tag number. Albacore were returned 

 to the water head first immediately after tagging. 



Labelle (1993b) described the albacore tagging op- 

 erations conducted in the south Pacific, and summa- 

 rized the corresponding release-recapture statistics. 

 Approximately 17,000 albacore were tagged and re- 

 leased from troll fishing vessels during 1986-92; the 

 season totals ranged from 815 to 6,524. As of Novem- 

 ber 1992, recaptures of 42 tagged albacore had been 

 reported; we obtained complete information on sizes 

 and dates of release and recapture for 27 of 42 alba- 

 core. Most of these 27 tags were returned by fishermen, 

 although some were detected during catch sampling at 

 canneries and landing sites. Recreational fishermen 

 also tagged and released an additional 3,646 albacore 

 along the south east coast of Australia during 1973-92 

 (Matthews and Deguara, 1992). As of November 1992, 

 14 of these had been recovered but only one tag recov- 

 ery record was sufficiently complete to allow inclusion 

 of the data. 



Data analysis 



Length-frequency analysis The MULTIFAN computer 

 program was used to estimate VB growth parameters 



from the length-frequency data under the assumption 

 that the modes in the data represent year classes. A 

 detailed description of the model is given by Fournier 

 et al. (1990) and the program is described in the 

 MULTIFAN 3 User's Guide and Reference Manual (Ot- 

 ter Research, 1991). MULTIFAN can incorporate spe- 

 cific structural hypotheses into models being fitted to 

 the length-frequency data. The simplest structural hy- 

 pothesis assumes that the mean lengths-at-age lie on 

 a VB growth curve and that the standard deviations of 

 length-at-age are identical for all cohorts. More com- 

 plex structural hypotheses can be tested to determine 

 if they provide a statistically significant improvement 

 in fit to the data. The more complex hypotheses tested 

 assume that the following processes can occur in the 

 population sampled: 



1 Sampling bias for the first cohort. This could result 

 from selectivity during the sampling process in- 

 duced by the fishing gear or the sampling method. 

 Size selectivity was assumed to apply only to the 

 first cohort and to decrease linearly with age until 

 fish reach the second cohort. 



2 Age-dependent standard deviation in length-at-age. 

 For some fish populations, variation in length-at- 

 age is not constant across cohorts. This hypothesis 

 allows the standard deviation of length-at-age to 

 increase or decrease linearly with age. 



3 Seasonally oscillating growth. Seasonal growth pat- 

 terns are known to occur in some fish populations 

 (Pauly and Gaschiitz, 1979). This process was in- 

 corporated into the growth model by adding two 

 parameters, one representing the magnitude of the 

 seasonal effect and the other determining the time 

 of the year at which growth is slowest because of 

 the seasonal effect. 



We systematically fitted models incorporating all pos- 

 sible combinations of the above structural hypotheses 

 and used likelihood ratio tests to identify the most 

 parsimonious model structure. Fitting and testing pro- 

 cedures were done automatically by MULTIFAN, al- 

 though some user-specified input was needed to en- 

 sure that the model exhibited stable behavior (see 

 Fournier et al, 1990, for explanations). 



Estimates of the VB growth parameters K and L„ 

 were obtained along with parameters for sampling bias, 

 age-specific standard deviations, and seasonal growth. 

 In the absence of information on the age of the first 

 age-class, MULTIFAN assumes that the VB curve 

 passes through the origin (i.e., t =0). Estimates of age- 

 at-length are based on this assumption. 



Analysis of vertebrae data The VB growth param- 

 eters were estimated from vertebral-ring-count data 

 and length data, under the assumption that ring counts 

 indicate total age (in years). We did not attempt to 



