Terceiro and Idoine- SRLCA application to Pandalus borealis survey data 



771 



12 3 4 5 



AGE 



Figure 15 



Von Bertalanffy growth curves for Gulf of Maine PtiHt/a/Msftorpalfs. 

 Upper curve (soliti line) is for parameters derived from age-length 

 data in Haynes and Wigley (1969); center curve (large dashed line) 

 is for growth parameters cited in Mclnnes(1986); lower curve (dot- 

 dashed line) is for parameters derived by SRLCA for northern shrimp 

 survey length-frequency data. 



age the length frequencies provided mortahty esti- 

 mates consistent with those produced by visual inspec- 

 tion of length modes. 



Discussion 



We subjected SRLCA to a fairly stern test by attempt- 

 ing to interpret a data set exhibiting variable recruit- 

 ment and growth patterns, and by using a broad ini- 

 tial parameter search space. As noted in the Monte 

 Carlo tests of SRLCA by Basson et al. (1988), these 

 variations in recruitment, and in mean length-at-age 

 (presumed variable growth rate, especially for the 

 abundant 1982 cohort) between cohorts, made inter- 

 pretation of the northern shrimp length-frequency 

 distributions difficult. The shape and proximity of the 

 assumed age-2 and -.3 modes frequently caused SRLCA 

 to interpret these modes as a single age-group, result- 

 ing in highest scoring parameters that provided posi- 

 tively biased estimates of growth rate. This problem 

 was most severe for the annual distributions and per- 

 sisted in the pooled length frequency, although the 

 increased amount of information in the pooled distribu- 

 tion did increase the effectiveness of the SRLCA ap- 

 proach, witli a "correct" interpretation of the data 

 available from the quaternary score ridge. Analysis of 



the pooled data in a truly sequential fashion, after the 

 projection matrix approach of Rosenberg et al. (1986) 

 and Basson et al. (1988), as a supplement to SRLCA 

 might help in alleviating these problems. 



This exercise demonstrated that the best objective 

 fit obtained by SRLCA does not necessarily provide 

 the best interpretation of the data, as with most of the 

 existing length-frequency distribution analysis meth- 

 ods. Pragmatically, we could not rely on SRLCA to pro- 

 vide a single set (or even region) of growth parameters 

 that yield both the highest parameter score and the 

 "correct" interpretation of the data, unless supple- 



