Stem et al Fish-habitat associations at edge of Oregon continental shelf 



543 



Euclidean distance as a measure of similarity and the 

 group-average clustering method (Pimentel 1979). 



To examine specific associations between fish abun- 

 dance and bottom-type characteristics, data were ex- 

 amined using canonical correlation analysis (CCA). 

 CCA maximizes correlations among two sets of vari- 

 ables while it minimizes correlations within sets 

 (Pimentel 1979). We used CCA to quantify associations 

 between abundances of non-schooling fish species (data 

 set 1) and bottom types (data set 2). Our primary goal 

 was to extract meaningful, natural associations be- 

 tween fishes and habitat factors potentially influenc- 

 ing their distribution and abundance. CCA estimates 

 these associations using four metrics (Pimentel 1979). 

 First, the canonical correlation measures the overall 

 association between the two data sets. Second, the 

 redundancy coefficient measures the amount of overall 

 variation in one data set as predicted by the other. 

 While the canonical correlation coefficient describes the 

 goodness-of-fit of the two data sets, which can be in- 

 fluenced by a single high correlation between one 

 variable in each data set, the redundancy coefficient 

 measures the extent of overlap in the variation of the 

 two data sets. Third, the variable loadings indicate 

 which variables are correlated on a particular axis. The 

 fourth metric, canonical variate scores, measures the 

 contribution of each sampHng unit (in this analysis, the 

 habitat patch) to the fish-habitat pattern depicted on 

 each axis. Canonical variate scores are derived for each 

 data set: scores for the habitat data indicate the rela- 

 tive cover of specific bottom types on each axis, while 

 scores for the fish data indicate the relative abundance 

 of specific fish on each axis. Canonical variate scores 

 derived from CCA represent a powerful way to mea- 

 sure the abundance of fish in reference to habitat type. 

 In essence, the method controls for the effects of sam- 

 pling across a range of different habitats, and thus in- 

 creased our power to detect meaningful spatial varia- 

 tion in fish abundance. 



Data for CCA were derived using observations of 

 habitat patches, which were discrete segments of uni- 

 form bottom type within each transect {n 524 segments 

 for all transects). For each habitat patch, the abun- 

 dances of 21 fish species were tabulated relative to the 

 summed total area (in m-) comprised by the habitat. 

 For mixed bottom types, the total patch area was ap- 

 portioned 80% to the primary substrate and 20% to the 

 secondary substrate. 



Results 



The six stations represented a wide variety of sub- 

 strates, ranging from shallow rocky ridges separated 

 by sand, to intermediate-depth cobble and boulder 



fields, to deep mud and pebble bottoms (Figs. 1,2). Sta- 

 tions 1 and 3 (shallow bank tops) were rocky ridges at 

 60-80 m depth separated by sand and boulder- filled 

 valleys; station 2 (bank saddle) was primarily mud with 

 interspersed cobble at 150-200 m; station 4 was mud, 

 ridge, and cobble at 145-175m; station 5 was mud at 

 250-340 m; and station 6 was boulder and cobble 

 grading into mud at 200-270 m. Because transects 

 were always run into the current to insure control- 

 ability of the vehicle, distance traveled along the 

 transects was not standardized, but was in the range 

 467-2367m {x length 1357m, SE 460m). 



Nested two-way ANO VA of transects and observers, 

 based on the relative abundances of all non-schooling 

 fish species summed, indicated significant differences 

 among stations (F 6.22, df5,18, P<0.01), but not 

 among observers (F 1.39, df 2,18, P>0.05), or in inter- 

 actions among observer transects and stations (F 0.48, 

 df 10,18, F>0.05). A Student-Newman-Keuls multiple- 

 range test separated the mean number of non-schooling 

 fish at stations into two subgroups: station 4, where 

 fish were most abundant at 2.09 fish/m-, and all other 

 stations, which ranged between 1.84 fish/m^ (station 

 6) and 0.31 fish/m^ (station 1). 



Species identified: Number and size 



We identified 38 taxa to species in our 1988 dives. This 

 represents a 23% increase over the 31 species identified 

 in 1987 (Pearcy et al. 1989). The increase was due 

 primarily to species that were uncommon, suggesting 

 that we identified most or all of the numerically im- 

 portant species on Heceta Bank. There were distinct 

 differences in taxonomic composition and abundances 

 between non-schooling and schooling fishes. About 89% 

 of the non-schooling fishes seen were identified to 

 species; fewer than 2% were not identified to family 

 or genus. All schooling fishes seen were Sebastes. Of 

 these, only 49% were identified to species; the re- 

 mainder were identified to genus only (Table 1). Most 

 of the schooling fish were small or juvenile fish that 

 we could not identify without voucher specimens. 



We counted 10,102 non-schooling fish, ranging from 

 3829 individuals of pygmy rockfish Sebastes wilsoni to 

 one individual in each of ten species (Table 1). School- 

 ing fishes comprised 22,470 individuals, over 50% of 

 which (12,820) were unidentified small Sebastes. The 

 most abundant identifiable schooling species was again 

 the pygmy rockfish, with 8390 individuals. The least- 

 abundant schooling species was widow rockfish 

 Sebastes entomelas with 20 counted. The total number 

 of fish schools seen (all species) was 145, ranging from 

 70 pygmy rockfish schools to one school of widow 

 rockfish. The number of individuals per school ranged 

 from about 10 to 330. 



