Bacheler and Shertzer: Estimating relative abundance and species richness from video surveys of reef fishes 
23 
Figure 6 
Relationship between the probability of a species being 
observed in a video segment and (A) its mean time in 
video or (B) the mean number of individuals of that 
species observed in each video. Included in the analysis 
were 90 reef fish species present in at least 10 videos 
in the National Marine Fisheries Service’s reef fish 
video survey conducted in the northern Gulf of Mexico 
in 2001-2002 and 2004-2007 as part of its Southeast 
Area Monitoring and Assessment Program. The solid 
black lines indicate fitted relationships from a gener- 
alized additive model, the dashed lines are 95% confi- 
dence intervals, and the tick marks on the x-axis show 
the distribution of values from the 90 species that were 
included in this analysis. 
be read is likely study-specific and would depend on 
the total number of videos to be read, the relative im- 
portance of rare species, the total resources available 
for video reading, and time constraints for video read- 
ing (Gledhill, 2001). Our results indicate that reading 
approximately 50 frames from each video may provide 
a reasonable compromise between costs and informa- 
tion gained, if one can accept that about 14% of species 
would be missed at each site compared with a reading 
of all frames in an entire 20-min video segment. 
As shown by our GAM results, behavioral character- 
istics largely determined how likely a species was to be 
observed or missed in a subset of video frames. Wheth- 
er a species is observed in a subset of video frames is 
almost entirely dependent on 2 behavioral character- 
istics: 1) the mean duration of time spent by each fish 
in the video viewing area and 2) the mean number of 
individuals present in each video. Fast-swimming and 
relatively infrequent fishes, such as tunas, mackerels, 
barracudas, and jacks, were the ones most likely to be 
missed and, therefore, tended to be underrepresented 
in estimates of species richness. These same taxa also 
had higher absolute CVs around indices of abundance 
than those for fishes like groupers and snappers that 
were observed more frequently. We also showed that 
CVs from observations of a fast-moving, schooling 
species (vermilion snapper) were more than twice as 
high as CVs for a slow-moving, nonschooling species 
(scamp), but the relative pattern of CVs was the same 
for both species. Clearly, researchers must carefully 
consider the behavior of their target species when de- 
signing a BRUVS sampling strategy with a MeanCount 
approach, for instance, by allocating significantly more 
video-reading effort if fast-moving, infrequently en- 
countered species are targeted. 
We estimated the proportion of species observed in 
a subset of frames compared with the number observed 
in all frames of a 20-min video segment, but note that 
reading all frames in a 20-min video segment likely 
underestimates all the species present at a site. For 
instance, Gledhill (2001) showed that approximately 
68% of reef fish taxa in the Gulf of Mexico that were 
observed in a continuous 60-min video segment were 
observed in analysis of a 20-min segment. Further- 
more, given the exclusively diurnal sampling in our 
study, nocturnal fishes were likely poorly detected, as 
were small, cryptic species (Collette et al., 2003; Smith- 
Vaniz et al., 2006; Williams et al., 2006). Therefore, our 
results (from an approach for which a subset of frames 
was read) should be interpreted as a reduction in spe- 
cies observed compared with results from reading of a 
20-min video segment, not a comparison with the true 
species richness at a site. 
Researchers could consider approaches that account 
for the fact that all video reading methods likely miss 
some reef fish species that are actually present at a 
site. First, occupancy or W-mixture modeling approach- 
es can estimate detection or capture probabilities sepa- 
rately from the underlying distribution or abundance 
of a species (MacKenzie et al., 2002; Royle, 2004), but 
multiple site visits may be necessary each year (Issaris 
et al., 2012) unless spatial aritocorrelation is modeled 
(Johnson et al., 2013). Second, if the emphasis is on es- 
