470 
Fishery Bulletin 108(4) 
occurrence, the species observed was probably the Pa- 
cific hagfish (Eptatretus stoutii) (Barss, 1993). 
All fish in the videotapes were counted and assigned 
an activity category (“active” or “inactive”). The two cat- 
egories were developed in order to analyze the videotape 
efficiently and quantitatively, and all fishes were placed 
into one of the two categories. Active fish were off the 
bottom (or temporarily in contact with the substrate), 
and inactive fish were in contact with the substrate 
(e.g., in contact with the seafloor or were occupying 
crevices). All flatfish were excluded from the activity 
analysis because of our definition of “activity.” We as- 
sumed that fish were counted only once and that the 
submersible did not influence the activity of fishes. 
When the ROV clearly affected the behavior of a fish, 
the activity of the fish observed before the ROV inter- 
ruption was used in analyses. 
Data analysis 
Relative abundance was determined for each taxon on 
a broad scale over all stations (2, 3, 4, 6, and 9) and all 
habitat types, and on a finer scale within each of the 
four primary habitat types (rock ridge, boulder, cobble, 
and mud) over all stations. Fish abundance was first 
normalized by dividing the abundance for a given taxon 
in a habitat patch by the area swept in that habitat 
patch. Relative abundance in a daytime habitat patch 
was matched up with a relative abundance in a night- 
time habitat patch of closest geographic proximity for a 
given taxon. These pairs were used to create the ranks 
in a Wilcoxon signed-rank test. For all taxa, 402 habitat 
patches were included in the analysis, and an average of 
59 habitat patches were compared for each taxon (many 
taxa were in the same habitat patch). This enabled us to 
estimate if relative abundance trends were consistent at 
both scales. For the finer scale, some habitat types (e.g., 
flat rock) did not contain sufficient relative abundance 
for analysis. 
The Wilcoxon signed-rank test was performed with 
S-Plus, vers. 3.2 software (TIBCO Software Inc., Palo 
Alto, CA) to determine significant differences in the ac- 
tivity of fishes during day and night (Ramsey and Scha- 
fer, 2002). This test involved estimating the percentage 
of fish active and inactive within each habitat patch for 
each taxon (raw abundance was used). These day and 
night pairs (percentage of fish active and percentage of 
fish inactive) were used to create the ranks in the test 
for a given taxon over primary habitat types. A total of 
398 habitat patches were analyzed for all taxa, with an 
average of 64 habitat patches compared for each taxon. 
We also used nonmetric multidimensional scaling 
(NMS) to examine associations between day and night 
fish abundance with depth and with primary habitat 
(McCune and Grace, 2002). PC-ORD software, vers. 
5.0 (MjM Software, Gleneden Beach, OR), was used 
with a Monte Carlo test and Sprensen distance mea- 
sure, starting with random configurations (Mather, 
1976). We restricted the NMS to taxa that showed 
significantly greater abundance during day or night in 
the Wilcoxon signed-rank test (P<0.05), and to those 
that showed a strong correlation (Pearson and Kendall 
correlation) with depth (P<-0.5 or >0.5 on the second 
axis) during trial NMS runs with all 31 taxa. A total 
of 11 taxa met these criteria. The final species (taxa) 
matrix included columns of log-transformed abundance 
and rows of sample units grouped by primary habitat 
for each dive during day and night (e.g., boulder-night - 
R534). All primary habitat types were used because 
the NMS determines correlation strength along an 
environmental gradient and does not require paired 
plots as in the Wilcoxon signed-rank test. The final 
environmental matrix included two quantitative vari- 
able columns: primary habitat types (flat rock, rock 
ridge, boulder, cobble, pebble, sand, and mud) and 
average depth (meters). Sample units greater than 3.0 
standard deviations were excluded from analysis. The 
final ordination had 166 iterations and 15 runs. This 
test enabled us to determine if marked differences in 
fish abundance were associated with depth and pri- 
mary habitat, and whether taxa showing significantly 
greater abundance during day or night were distrib- 
uted similarly on the bank. 
Results 
A total of 29,787 individual fish were counted on the 
ROV transect videotapes. During the day, we observed 
an average of 207 fishes per hectare, and at night we 
observed a lower average of 141 fishes per hectare. 
Fish taxa in greatest abundance were from the genus 
Sebastes. Dominant taxa (pygmy and Puget Sound rock- 
fish, and unidentified juvenile rockfish) showed the 
largest differences in relative abundance between day 
and night (Fig. 3). Across all stations and primary 
habitat types, and within at least one primary habitat 
type, eight taxa showed significantly greater abundance 
during the day (P<0.05) and five taxa exhibited sig- 
nificantly greater abundance during the night (P<0.05, 
Table 1 , Fig. 3). Three taxa were found to be signifi- 
cantly greater in abundance during day (kelp greenling 
and unidentified mottled poacher [Agonidae]) and night 
(redstripe rockfish [S. proriger ] ) only in specific primary 
habitat types (Table 1). Several taxa showed apparent 
differences in abundance, but sample sizes were too 
small for statistical significance in the paired Wilcoxon 
signed rank test (e.g., kelp greenling). Harlequin rock- 
fish ( S . variegatus) was the only species we regularly 
encountered exclusively at night (darkblotched rockfish 
were rare but were also seen only at night), whereas 
kelp greenling were encountered only during the day 
at shallow depths. 
NMS analysis of distribution 
The NMS analysis showed significant correlations 
(P=0.03) among taxa, depth, primary habitat, and day 
and night (Fig. 4). The ordination explained 78% of the 
variation with an acceptable stress value (a lower value 
