568 



Fishery Bulletin 100(3) 



o 



CD 



00 



in 



CD 



in 



CM 



in 



o 



CD 



00 



in 



CD 



in 



in 



c\j . 



in 



Estimated spatial trend in species diversity 



-170 -160 -150 -140 



Estimated spatial trend in total CPUE (log scale) 



-130 



Bering Sea 



-170 



-160 



-150 



Longitude ("W) 



-140 



-130 



Figure 5 



Spatial trends in species diversity (Shannon-Wiener index) and total CPUE. The trends are based on pre- 

 dictions from the best regression models for diversity and total CPUE (see Fig. 4 and Fig. 6). 



gested by a substantial overlap in the location of strata in 

 the ordination diagram. A three-way analysis of variance 

 for each of the three axes by depth stratum, area, and year 

 indicated highly significant differences for all three axes 

 between depth strata and geographic areas, but weak and 

 nonsignificant differences among years (Table 5), suggest- 

 ing a relatively stable species composition from 1984 to 

 1996. Residuals from the ANOVA for axis 1 and 3 were 

 approximately normally distributed and showed no appar- 

 ent violations of the ANOVA assumptions. Residuals for 

 axis 2 had a very long-tailed distribution, but an analysis 

 of variance based on ranks yielded very similar results; 

 therefore the conclusions appear to be robust. 



Ordinations of individual hauls within each year con- 

 firmed that depth and alongshore distance explained 

 much of the variation in species composition within years. 

 The species composition in each year was effectively sum- 



marized by ordinations in five dimensions with Kruskal's 

 stress values that were very close to 0.1 in all years (axes 

 1-5 in Table 6). The first axis of all five ordinations ac- 

 counted for A2-Al'7c of the variation in species composi- 

 tion. This axis, as well as most of the other axes, was most 

 strongly related to the depth gradient in all years (Table 6). 

 Alongshore distance appeared to explain a relatively 

 small proportion of the variance in species composition. 

 However the pseudo-coefficients of partial determination 

 for alongshore distance typically increased substantially 

 if Julian day was excluded from the model because of high 

 correlations between Julian day and alongshore distances. 

 Regressions of the axes on depth and alongshore distance 

 alone resulted in pseudo-r^ values that were very close to 

 those from models that included Julian day. Furthermore, 

 residuals from these models were not significantly related 

 to Julian day for any year (P>0.5) based on linear regres- 



