840 



Fishery Bulletin 100(4) 



E 

 < 



Depth (m) ^ ^ 



Estimated amount 

 50-500 m. 



Meuter (1999). A different measure 

 of similarity, based on converting 

 the data to presence-absence binary 

 form and computing the proportion 

 of nonzero values held in common, 

 has also been used (Krebs, 1989). 

 However, preliminary analyses indi- 

 cated very little difference between 

 the Bray-Curtis and binary similar- 

 ity measures (r=0.973). Therefore, 

 the Bray-Curtis similarity measure 

 was used in all subsequent multi- 

 variate analyses. 



Ordination techniques used in in- 

 direct gradient analysis were prin- 

 cipal components, detrended corre- 

 spondence analysis, and multidimen- 

 sional scaling (Rogers and Pikitch, 

 1992; Mahon et al, 1998; Meuter, 

 1999). Multidimensional scaling 

 (MDS) has the advantage of not 

 requiring an assumption about the 

 underlying response model and has 

 been shown to be robust to different 



relationships of species abundances and environmental 

 gi-adients (Minchin, 1987; Meuter, 1999). In our study we 

 used MDS in conjunction with the Bray-Curtis similarity 

 measure, similar to previous analyses (Field et al., 1982; 

 Meuter, 1999). The first three dimensions from the MDS 

 were extracted, plotted, and correlated with suspected 

 gradients, which were then used to numerically define as- 

 semblage groupings in the data. 



Species groupings or assemblages were also determined 

 by partitioning cluster analysis, rather than using other 

 commonly employed hierarchical cluster analyses (e.g. 

 Rogers and Pikitch, 1992; Weinberg, 1994; Mahon et al, 

 1998). Hierarchical cluster techniques result in dendro- 

 gram trees whose shape and structure depend largely on 

 the division and linking methods used, frequently result- 

 ing in little similarity among the many methods (Johnson 

 and Wichern, 1992; Ripley, 1996). Results from this type of 

 analysis are often reported for a single method, indicating 

 results consistent with the analyst's expectations, while 

 neglecting to report the range of variability produced by 

 the other alternative hierarchical methods. A more objec- 

 tive method of determining groupings is to use a partition- 

 ing technique such as the classical ^-means algorithm 

 (Hartigan and Wong, 1979). where the number of groups 

 (/;) is specified a priori and a single solution to the group- 

 ing structure is determined. 



For our partitioning analysis we used a more robust 

 variant of the ^-means method, which is termed ^-me- 

 dians (Kaufman and Rousseeuw, 1990). The /?-means 

 and ^-medians methods rely on the minimization of the 

 (dis)similarity between cluster centers and their mem- 

 bers. Specifically, the /f-means algorithm minimizes the 

 squared (dis)similarities, and /f-medians minimizes the 

 untransformed (dis)similarities, resulting in a measure 

 that is less sensitive to extreme values (Kaufman and 

 Rousseeuw, 1990; Ripley, 1996). In our study a range of ^'s 



>j?> o Latitude 



Figure 2 



of habitat (km) at half degree latitude intervals and depths of 



or cluster numbers was evaluated and the best grouping 

 structure for the data was determined according to the 

 highest average silhouette measure (Rousseeuw, 1987). 

 To detect misclassifications, the final rockfish assem- 

 blage structure was compared to the several dimensions 

 obtained from the MDS ordination analysis and to our 

 mapped CPUE distributions. 



Results 



There are a number of large-scale bathymetric features in 

 the Eureka, Monterey, and Conception INPFC areas that 

 may influence the distribution and abundance of shelf and 

 slope rockfish species. For example, the Mendocino Escarp- 

 ment is a large fracture zone that forms a huge submarine 

 ridge near Cape Mendocino that extends nearly 2500 km 

 westward into the Pacific Ocean and that measures 100 

 km across at its widest point. Well to the south of the 

 escarpment are a number of large submarine canyons in 

 the region of Monterey Bay and Point Sur These subma- 

 rine features, including Monterey and Sur Canyons, result 

 in a coastal bathymetry characterized by limited shelf 

 area and rapidly increasing depth. Farther south, in the 

 vicinity of Point Buchon, is a large offshore area that rises 

 to 430 m depth, i.e. the Santa Lucia Bank. Lastly, Point 

 Conception divides zoogeographic provinces and forms the 

 southern boundary of the area of this study. 



Our restricted analysis of the bathymetry at 0.5°-lati- 

 tude intervals indicates that variability in depth profiles 

 reflects some of the important features described above. 

 The amount of habitat in each 50-m depth interval seems 

 to peak in the 100 to 150-m depth range for latitudes 

 greater than 37.0°N (Fig. 2). The observed decrease in 

 the total amount of habitat at 40.0°N, especially in waters 

 shallower than 200 m, is due primarily to the Mendocino 



