20 
Fishery Bulletin 1 13(1) 
A 
C 
Fsgyre 2 
MeanCount, which is calculated as the mean number of in- 
dividual fish observed in a series of frames over a viewing 
interval, of a single sampling event (i.e., a 20-min video seg- 
ment) in the northern Gulf of Mexico as a function of the 
number of video frames that were read for (A) red snapper 
C Lutjanus campechanus), (B) vermilion snapper (Rhombop- 
lites aurorubens), and (C) scamp (Mycteropei-ca phenax). In 
each panel, the heavy solid line represents the mean of 1000 
bootstrap replicates, the lower dotted line represents the 5 th 
percentile, the upper dotted line represents the 95 th percen- 
tile, the thin solid line is a single bootstrap iteration, and 
the filled circle on the right is the true MeanCount across all 
1201 frames that were analyzed for this study. 
300% increase in frames read), CVs only declined 
by approximately 50% for all 3 species. 
In all cases, the relationship between mean 
CVs and number of frames read was described 
well by power functions (coefficient of determina- 
tion [r 2 ]>0.99; P<0.0001 for F-statistics). Among 
the 3 species, the estimated relationships show 
differences in scale (a') but similar rates of de- 
cline ( b ). The mean estimates of parameters and 
standard errors of the mean (SE) were a'=1.324 
(SE 0.004) and b=-0.525 (SE 0.001) for red snap- 
per, a'=1.585 (SE 0.004) and 6=-0.524 (SE 0.001) 
for vermilion snapper, and q'= 0.981 (SE 0.004) 
and b=- 0.525 (SE 0.001) for scamp. Because the 
rates of decline were similar across species, the 
same number of frames could be read for each if 
the goal is to achieve a proportional reduction in 
CV from each species’ maximum (which occurred 
at F= 1). However, if the goal is to achieve a par- 
ticular CV, then each species would require a dif- 
ferent number of frames read. For example, a CV 
of 0.4 would require F=71 for red snapper, F=119 
for vermilion snapper, and F = 37 for scamp. 
Species richness 
For illustration, we chose 3 example videos with 
varying levels of species richness to show general 
patterns in estimates of species richness. In these 
example videos, the estimated species richness 
(number of species observed) increased asymptoti- 
cally as more frames were read (Fig. 4). Although 
these 3 videos were chosen simply as examples, 
they show that the estimates of species richness 
increased stepwise as the number of frames read 
was incremented. That is, with the inclusion of 
each additional frame read, the species count ei- 
ther remained the same if that frame contained 
no new species or it increased by the number of 
new species observed. In these examples, the me- 
dian estimate captured 100% of the species pres- 
ent by around 50 frames for the videos with 3 or 7 
species observed but not until around 100 frames 
for the video that contained 17 species (Fig. 4). 
Across all 1543 videos, the proportion of spe- 
cies observed increased with sampling intensity. 
Most species (median proportion=0.75) were ob- 
served in each video when only 25 of the 1201 
frames were read, and the median proportion in- 
creased when 50 frames (0.86), 100 frames (0.95), 
or 200 frames (0.99) were read (Fig. 5). On the ba- 
sis of the interquartile range, however, there was 
substantial variability among videos, particularly 
when fewer than 100 frames were read (Fig. 5). 
When 100 or more frames were read, the vari- 
ability was lower but some rare species in some 
videos were still missed. 
The GAM explained 87.2% of the deviation in 
the probability that a species would be observed 
