844 



Fishery Bulletin 100(4) 



A general comparison of interspecific relationships, 

 based on depth-latitude distributions, can be gathered by 

 inspection of a plot of CPUE-weighted depth-latitude cen- 

 troids (Fig. 11). From results presented in Figure 11, some 

 species (e.g. copper \S. cam-inus] and brown [S. auricula- 

 tus] rockfish) appear to have very similar depth-latitude 

 distributions. Aurora and blackgill rockfish are clearly 

 deep-water southern species, whereas halfbanded (S. semi- 

 cinctus) and vermilion (S. miniatus) rockfish are southern 

 species found principally in shallow water. Although this 

 plot is useful for identifying related species by their aver- 

 age distribution in space, it does not fully represent actual 

 joint co-occurrences and assemblage relationships, as does 

 a complete multivariate community analysis. 



The results of the multidimensional scaling of the 

 fourth-root transformed CPUE data, with the Bray-Cur- 

 tis similarity measure, revealed highly significant cor- 

 relations of the first three dimensions with mean depth 

 (7-0.90, P<0.0001), total CPUE"-'^ (r=-0.82, P<0.0001), 

 and mean latitude (r=-0.74, P<0.0001), respectively. How- 



ever, these first three dimensions accounted for only 39% 

 of the total variance in the data. 



The ^-medians partitioning analysis was used to exam- 

 ine a range of ^'s or cluster numbers, effectively predefin- 

 ing the number of distinct assemblages. The best fits were 

 determined by the highest average silhouette measures of 

 0.143 and 0.140 for /; = 8 and ^ = 4, respectively (Table 4). 

 According to Kaufman and Rouseeuw (1990) an average 

 silhouette measure less than 0.25 does not indicate any 

 substantial structure in the data. Despite the low aver- 

 age silhouette measurements, the suggested gi-oupings of 

 four and eight clusters were used as an initial guideline 

 in determining distinct species assemblages. The species 

 groups defined in Table 4 by a division into four clusters 

 (A, B, C, and D), and further delineated in Figure 11, show 

 that these assemblages generally follow depth and the 

 latitude distributions of their member species. 



The first three dimensions from the multidimensional 

 scaling analysis, along with the k = 8 cluster divisions, are 

 shown in Figure 12. The eight clusters comprising the 26 



