Anderson and Yoklavich Habitat association of deepwater demersal fishes off central California 



171 



plots (i.e., \og(variance) versus \og{mean)). Data were 

 generally right-skewed and had a positive variance- 

 mean relationship. The slope of the Taylor power plot 

 was used to optimally decouple variance from mean by 

 raising the data to the power of ((2-slope)/2) (McArdle 

 et al., 1990). Consequently, species abundance data were 

 ^y■o ir.^ transformed, total abundance was transformed by 

 loglO(.r-i-l), and a square root (.v" ''I transformation was 

 applied to species richness. To examine broad-scale 

 relationships between fish species and benthic habitat 

 variables, we ran a canonical correlation analysis on 

 the transect-level data matrix and then plotted the total 

 structure coefficients of the fish in habitat space. The 

 standardized redundancy output values of the model 

 were used to measure the amount of variation for both 

 fish species and benthic habitat variables. 



To examine intermediate-scale relationships between 

 fish species and benthic habitat variables, densities of 

 fishes per patch types were examined. However, be- 

 cause all patch types were not equally available, we 

 also standardized patch-use relative to habitat avail- 

 ability (patch selectivity) by subtracting proportional 

 occurrence of each patch type from the proportional 

 abundance for each species. Here, a positive association 

 with a patch type revealed that more individuals were 

 found in that patch type than would be expected given 

 random habitat use (i.e., no selectivity). Conversely, a 

 negative association revealed that fewer individuals 

 were found in that patch type than would be expected 

 by random habitat use. Finally, because microhabitat 

 availability was not measured independently of fish 

 presence, microhabitat use by fishes was restricted to 

 graphical presentation. 



Results 



Seafloor composition 



We sampled 11.15 linear km of seafloor within the 12 x 

 10 km survey region, using submersible strip-transect 

 methods. At broad-scales, benthic habitat variables were 

 grouped a posteriori in order to reliably distinguish 

 hard, mixed, and soft strata (Fig. 2). Hard stratum 

 comprised patchy "high-relief outcrops" of rock, boul- 

 ders, and sand. In contrast, mixed stratum comprised 

 "low-relief outcrops" of cobbles and mud. Soft stratum 

 comprised "homogeneous mud." The three broad-scale 

 habitat strata also varied in their depth distribution, 

 and strata and depth were strongly collinear. High-relief 

 outcrops were generally shallower (60-100 m) than low- 

 relief outcrops (90-150 m), and although homogeneous 

 mud occurred in most depth ranges, it was the only 

 stratum surveyed in deep offshore locations (80-260 m). 

 Benthic habitat variables within each of the three strata 

 were also strongly collinear. For example, rock always 

 co-occurred with boulders and sand, forming complex 

 high-relief outcrops in shallower water (i.e., <100 m). 

 Therefore, if a species was correlated at broad spatial 

 scales with high-relief outcrops, differentiating the rela- 



tive importance of substratum composition, depth, or 

 some corequisite would be problematic. 



Variability in intermediate-scale habitat also was dis- 

 cernible (Fig. 3). Five substrata (rock, boulders, cobbles, 

 sand, and mud) were recorded during this survey, which 

 at intermediate scales were present in 21 of 25 possible 

 paired "substratum patch types" fall but mud-sand, 

 cobble-sand, sand-cobble, or sand-mud patches types 

 were recorded). However, the proportional availabil- 

 ity of these patch types differed between strata. For 

 example, hard strata contained the highest number 



