422 
Fishery Bulletin 115(3) 
near to and far from the horizon in video recordings 
of fish aggregations and habitat. An on-screen display 
system recorded and superimposed time, date, ROV 
heading (direction), and ROV depth on the video taken 
with the cameras. The ROVs also had high-resolution 
digital cameras that captured still images of fish and 
habitats within the study area. The still cameras were 
mounted on the ROV in a fixed, downward-looking 
orientation for images of habitat cover. Both cameras 
were equipped with parallel lasers (10 cm) to calculate 
scale. Two 250-W halogen lights (DeepSea Power & 
Light, San Diego, CA) were mounted on top of the ROV 
tilt platform and provided illumination for the video 
cameras on the Phantom, and the Mohawk ROV had 
two 3700-lm SeaLite Sphere 3100 LED lights (DeepSea 
Power & Light). Water clarity and natural light, how- 
ever, usually allowed visibility in excess of 20 m. When 
available, an SBE 39 temperature and depth recorder 
(Sea-Bird Scientific, Bellevue, WA) was attached to the 
ROV for each dive. 
A statistically rigorous sampling protocol was used 
to select the ROV survey sites at Pulley Ridge. In Ar- 
cGIS, vers. 10.1 (Esri, Redlands, CA), a grid of blocks, 
each lxl km, was overlaid on maps created with mul- 
tibeam bathymetric imagery collected by NOAA 7 (Fig. 
1). Blocks were selected randomly to be surveyed quan- 
titatively by the ROV over the 4 years, and the pooling 
of blocks for selection targeted both the Pulley Ridge 
HAPC and areas adjacent to the HAPC that appeared 
to be mesophotic habitat from the bathymetric maps. 
Areas outside the HAPC had been mapped previously, 
but the bathymetric maps have not yet been verified 
by direct observations, or ground-truth; therefore, ar- 
eas interpreted as hard-bottom habitat from the bathy- 
metric data were used in the selection of blocks. Once 
a block had been examined, it was not resampled in 
subsequent years. Each dive of the ROV lasted approx- 
imately 3-4 h during daylight hours and covered an 
average distance of 1.85 km (standard error [SE] 0.11). 
The direction of each dive within a block was selected 
haphazardly on the basis of a flip of a coin and the 
maneuverability of the ship, which is affected by wind 
and current, but the direction was not altered to target 
grouper pits. 
Video reading 
All fish were counted and identified in each encoun- 
tered grouper pit, including species that were both in- 
side the pit and swimming in the water column above 
(1-3 m) the pit. Individual fish were identified to the 
lowest taxonomic level possible, and fish counts for 
each taxon were made by using a tally system. Still 
images of single frames of video were used to identify 
and count fish when multiple species were present and 
when areas had high fish abundance. If confident iden- 
tifications could not be made, individuals were recorded 
as unknown. Random segments of video were analyzed 
by a second reader to confirm identification of fish and 
accuracy of the primary reader’s counts. Counts for 
large schools of fish (>100 individuals) were estimated 
by counting a group of 25 fish and then extrapolating 
that count for the size of the entire school. To avoid 
recounting fish, unique color patterns, body markings, 
and attraction behaviors (i.e., schooling of fish around 
the ROV) were noted. Fish abundances were recorded 
for each taxon observed in every grouper pit. Because 
pits were of relatively similar sizes, averaging ~10 m 
in diameter (as measured from high-resolution, multi- 
beam bathymetric imagery) and 1-2 m in depth, fish 
abundance per pit, rather than density, was used. Cole- 
man et al. (2010) found no relationship between pit di- 
ameter and either fish density or species abundance in 
the Madison-Swanson and Steamboat Lumps MPAs in 
the northeastern Gulf of Mexico. 
Analyses of multivariate fish communities 
Multivariate analyses were conducted by using PRIM- 
ER, vers. 6 (PRIMER-E, Auckland, New Zealand) to 
compare fish communities in the grouper pits. Each 
grouper pit was defined and characterized by the fol- 
lowing 4 factors: year, predator presence or absence, 
region, and HAPC. Year indicated the year the grouper 
pit was sampled: 2012-2015 (Fig. 2A). Red grouper and 
lionfish are the 2 top-level predators that inhabit the 
grouper pits. To test the effect of predator presence or 
absence on community structure, grouper pits were cat- 
egorized as 1) having either no predators (no lionfish 
or red grouper), 2) lionfish only, 3) red grouper only, 
or 4) both (red grouper and lionfish present) (Fig. 2B). 
Although in some cases on shallow reefs lionfish have 
been observed to move more than 1 km (Akins et al., 
2014; Tamburello and Cote, 2015), both lionfish and red 
grouper are known for their site fidelity (Coleman et 
al., 2010; Jud and Layman, 2012; Bachelor et al., 2015). 
Coleman et al. (2010) examined the movement patterns 
of red grouper in pits specifically and found that they 
exhibit high site fidelity, remaining in the same pit for 
long periods of time (>1 year). Other predators around 
the grouper pits, such as other grouper species and 
species of snapper, are more roving predators. The re- 
gion factor indicated the location of the grouper pits in 
relation to the geological features of the Pulley Ridge 
MCE, primarily on the basis of bathymetric maps. Four 
geological regions were used to categorize the location 
of each grouper pit: main ridge, off main ridge (area 
east of the main ridge), central basin, and west ridge 
(Fig. 2C). The HAPC factor indicated whether the pit 
was located inside or outside the HAPC. 
Four multivariate routines in PRIMER were em- 
ployed to examine fish communities by using the fac- 
tors described previously. They were nonmetric mul- 
tidimensional scaling (MDS), analysis of similarity 
(ANOSIM), similarity percentages (SIMPER), and bio- 
diversity indices (DIVERSE) routines. For these analy- 
ses, taxa that composed less than 1% of the total abun- 
dance were removed to minimize the disproportionate 
effect they can have on the data. Data were averaged 
by factor and fourth-root transformed — a calculation 
