Ross et al.: Fish species associated with shipwreck and natural hard-bottom habitats 
49 
of biota and habitats. The color video cameras attached 
to the ROVs had scaling lasers (10-cm spacing) used to 
estimate total length (TL) of fish. During transect sur- 
veys, the cameras were positioned to record directly in 
front of the ROV and were set on wide angle (or near 
wide angle). The video cameras recorded continuously 
throughout the ROV dives (whether the ROV was mov- 
ing over transects or was stationary on bottom), and 
digital still images were taken frequently to augment 
video collection. 
Habitat definitions 
A main objective was to determine to what degree 
fishes were associated with general habitats on a large 
scale; therefore, habitat definition was reduced to 2 
broad, relatively simple types: 1) soft substrata (SS) 
of sand or mud — relatively flat substrata and with 
few structuring features aside from gravel, burrows, 
depressions, and animal tracks; and 2) artificial (ship- 
wreck) substrata and natural hard bottom (AS/NHB), 
which included World War I-era shipwrecks with sub- 
stantial vertical profile and one site with natural hard 
bottom (consolidated mud, ledges, and boulders). Ad- 
ditional habitat metrics included bottom depth and 
environmental data recorded by the CTD instruments 
mounted on the ROVs. 
Video analysis: community and habitat association 
A preferred method for documenting fauna in complex 
habitats, visual observations (here based on ROV-col- 
lected video) were used to describe the fish communi- 
ties and associated habitats at the 9 study sites. Tracks 
of ROV dives were processed initially to conservatively 
remove erroneous tracking data (location points) as 
described by Quattrini et al. (2012). To determine com- 
munity structure and habitat associations of fishes at 
sites, much as described in Ross and Quattrini (2007), 
videos from each dive were viewed multiple times for 
habitat classifications (see “Habitat definitions” sec- 
tion) and for identification (to the lowest possible taxa) 
and enumeration of fishes by time of observation. Video 
segments were designated when the ROV stopped or 
started movement, when the video quality changed, or 
when the habitat changed. Depth was recorded by the 
ROV-mounted SBE 911plus for every time segment. 
Unusable video (out of focus, too far off bottom, because 
of malfunction, sediment clouds) was removed from the 
data set. 
Species composition and relative abundances (fish 
counts) were determined from the wide-angle video and 
were compared within each of and between the 2 habi- 
tat types. To compare abundances of all species within 
a habitat type, relative abundances were calculated in 
percentages as the number of individuals per taxa per 
habitat type divided by the total number of individuals 
observed per habitat type and then multiplied by 100. 
For comparisons between habitat type, analysis was 
restricted to benthic fishes identified to at least fam- 
ily level and with overall abundances >2. Occurrence 
of at least 2 individuals allowed for the possibility of 
a taxa occurring in both habitat types. Relative abun- 
dances by habitat type were calculated for each taxon 
by dividing the number of individuals in a particular 
habitat type by the total number of individuals of the 
same species from both habitat types and multiplying 
the result by 100. 
Multivariate analyses were conducted in PRIMER 6 
and PERMANOVA+ (PRIMER-E Ltd., Ivybridge, U.K.) 
(Clarke and Warwick, 2001; Clarke and Gorley, 2006; 
Anderson et al., 2008) to determine differences in ben- 
thic fish assemblages between habitat types. Sample 
units were the numbers of each species per habitat type 
(SS or AS/NHB) per ROV dive; samples with no species 
present were removed from the data set. Because tran- 
sect times were variable, abundances of species were 
standardized per sample by dividing the number of 
individuals per species by the total number of fishes 
per sample. Standardized abundances were fourth-root 
transformed to down-weight the abundant species in 
relation to rare species. The Bray-Curtis similarity 
coefficient was used to calculate similarities between 
samples, and on the basis of the resulting similarity 
matrix a nonmetric multidimensional scaling ordina- 
tion (MDS) plot and a dendrogram with group aver- 
age linking were created. One-way analysis of similari- 
ties (ANOSIM) and post-hoc multiple comparison tests 
were used to determine whether there were significant 
differences between fish assemblages in the 2 different 
habitat types. Similarity percentage (SIMPER) analy- 
sis was used to determine which species contributed to 
the dissimilarities among habitat types. 
Results 
On the 9 study sites (depths of 42-126 m), 14 ROV 
dives were completed, 9 dives in September 2012 and 5 
dives in May 2013 (Table 1, Fig. 1), resulting in 84.4 h 
of usable video data on hard-bottom (AS/NHB) habi- 
tat and 16.5 h of video data on soft-bottom (SS) habi- 
tat. Soft-bottom habitat was observed exclusively with 
video collected during the dive at the shallowest site 
(Table 1, Fig. 1); however, because only 3 specimens of 
unidentified skates were observed during this dive, it 
made little contribution to our study. Although ship- 
wrecks and natural hard bottom were the focus of the 
remaining dives, soft-bottom habitat surrounding those 
hard-bottom habitats was also surveyed during these 
dives. 
In September 2012, mean bottom temperatures 
varied about 2.5°C across the study sites; the coldest 
temperatures (means: 11.9-13.0°C), lowest salinities 
(means: 33.1-34.8), and highest DO (means: 4. 0-4. 5 
mL/L) occurred at the shallower sites (depths of 42-81 
m) (Table 2). At each of the 5 deeper sites (depths of 
84-126 m), bottom temperatures (means: 14.2-14.5°C), 
salinities (35.6-35.8), and DO (3. 7-4.0 mL/L) were 
similar to each other. In May 2013, little variation was 
