542 



Fishery Bulletin 90(3). 1992 



moving syncronously in the same direction). Non-polar- 

 ized aggregations or solitary individuals were consid- 

 ered "non-schooling." Data w^ere collected by contin- 

 uous audio tape recordings of the observer during 

 transects, continuous video records (also including 

 audio, time, and date), and 35 mm still photographs 

 automatically triggered every 30 seconds. We used a 

 PhotoSea 1000 35 mm still camera and a PhotoSea 2000 

 video camera, both on fixed mounts outside the vehi- 

 cle. The video camera was mounted on the starboard 

 bow of the submersible and recorded a field of view that 

 partially overlapped that of the observer within the 

 submersible. The audio track of the videotape recorded 

 the observers comments which allowed real-time inte- 

 gration of fish observations and bottom-type descrip- 

 tions (see below). Visibility always extended at least 

 to the limits illuminated by the lights (i.e., ~3m or more 

 except where limited by topography). Immediately 

 following each dive, data were entered by computer 

 into a relational database system and verified against 

 the audio tapes. 



We tried to minimize inherent biases of submersible 

 studies as suggested by Ralston et al. (1986), such as 

 fishes not seen or unidentified, diurnal variability, and 

 effects of vehicle on fishes. Through a detailed analysis 

 of fish and bottom-type observations recorded in the 

 continuous video coverage of each transect, several 

 observer-related factors affecting data collection were 

 discovered. First, the diving observer usually noted 

 fishes first, then bottom type. When fishes were pres- 

 ent coincidentally with a substrate change, fish records 

 were frequently correlated with the wrong bottom 

 type. Second, observers tended to record substrate 

 types based upon larger (high-relief) features rather 

 than small (low-relief) ones, even when the smaller ones 

 were preponderant. Apparently, boulders impressed 

 observers more than cobble or mud, even when the 

 latter were most abundant. Neither of these sources 

 of error was intuitively obvious or suspected. If left 

 uncorrected, these errors would have changed the 

 apparent fish-substrate associations. 



Due to these inherent biases, we extracted data on 

 bottom types from the videotape record of each tran- 

 sect. In order to standardize any bias in the evaluation 

 of bottom types, a single observer (BT) reviewed all 

 videotapes. Dominant substrates were categorized 

 using a two-code combination of seven possible cate- 

 gories: mud (code M), sand (S), pebble (P, diameter 

 <6.5cm), cobble (C, >6.5 and <25.5cm), boulders (B, 

 >25.5cm), flat rock (F, low vertical relief), or rock 

 ridges (R, high vertical relief). Substrate was noted as 

 either "primary" if it covered at least 50% of the area 

 viewed (the first code), or "secondary" if it covered 

 more than 20% of the area viewed (the second code). 

 For example, a mud-boulder bottom type (code MB) 



consisted of at least 50% cover by mud with at least 

 20% cover by boulders. In contrast, a mud bottom (MM) 

 consisted of >80% cover by mud. 



We defined each transect segment of uniform bot- 

 tom type as a "habitat patch." Transects within sta- 

 tions were therefore represented by a series of habitat 

 patches defined by the frequency of substratum change 

 along a transect. As a result, the size of habitat pat- 

 ches varied both within and among transects in con- 

 junction with the area of uniform bottom types. The 

 average habitat patch measured 150.8 m^ (SE 15.4 m^, 

 n 524). 



Data analysis 



Although data were collected on all observed fish, data 

 analysis focused on the distribution and abundance of 

 non-schooling fishes rather than schooling fishes, 

 because data for the former were more reliable. First, 

 due to the lack of a manipulator on the submersible, 

 we were unable to collect schooling fishes, which were 

 typically small and unidentifiable, to obtain voucher 

 specimens for positive identification. Second, school- 

 ing species were generally more abundant above the 

 bottom in midwater and were not common in the tran- 

 sect path. 



We tested for statistical differences among stations 

 and observers in non-schooling fish abundance using 

 a nested two-factor analysis of variance (ANOVA). 

 Thirty-minute transect segments served as nested 

 replicates. Sample variances were examined for 

 homogeneity using Bartlett's test (Sokal and Rohlf 

 1981) prior to using the ANOVA. Because the raw data 

 were heteroscedastic, the analysis examined the log- 

 transformed total abundance of non-schooling fish 

 per m^. 



To examine the variation in fish assemblages among 

 transects, data were analyzed using principal compo- 

 nent analysis (PCA). The PCA was an R-mode analysis 

 of the variance-covariance matrix based on the log- 

 transformed abundance of non-schooling fish per m^. 

 By definition, the axes examined in PCA are statistical- 

 ly independent of on another (Pimentel 1979). Rare 

 species were eliminated from analysis by selecting only 

 species present on at least 10 of the possible 36 

 transects. A total of 30 taxa met this criterion and were 

 used in the analysis. 



To examine the overall similarity of fish assemblages 

 occurring on different substrates, data were analyzed 

 using hierarchical cluster analysis. The analysis was 

 limited to 21 species which were present on at least 

 12 of the possible 36 transects. The data in this analysis 

 were the log-transformed total number of individuals 

 per m^ of each species that occurred on each substrate 

 combination. A dendrogram was constructed using 



