Merritt et al.: BotCam: a baited camera system for nonextractive monitoring of bottomfish species 
59 
were defined on the basis of intersecting substratum 
(bottom) hardness and slope: 1) hard bottom-high 
slope (HB-HS); 2) hard bottom-low slope (HB-LS); 3) 
soft bottom-high slope (SB-HS); and 4) soft bottom- 
low slope (SB-LS). High slope values were considered 
to be 20 degrees or greater and hard substrata had 
backscatter values equal to or greater than 41 on a 
scale of 0-100 (actual maximum measurement was 
92). The sampling locations were randomly selected 
within these four habitat types and weighted towards 
the preferred bottomfish habitat. A total of 38 sites 
were sampled on HB-HS, 14 on HB-LS, 17 on SB-HS, 
and 13 on SB-LS. In this way greater replication was 
performed where fish densities were expected to be 
higher and replication was lower where few or no fish 
were expected to be found. Adjacent sampling locations 
were no closer than 200 m and to avoid cross influence 
of the bait, no two adjacent sites were sampled on the 
same day. 
The BotCam system was set to begin recording after 
its release from the boat but before its arrival on the 
bottom. For each deployment, the recording period was 
between 45 and 60 minutes. The bait consisted of equal 
parts of ground squid and mackerel, and the volume of 
bait used for each deployment was standardized to ap- 
proximately 1 liter. This mixture was designed 1) to be 
similar to what bottomfish fishermen typically use on 
their rigs; 2) to provide multiple types of scent; and 3) 
to provide food similar to the natural diets of the “deep 
7” which include both fish and cephalopods (Haight et 
al., 1993b). 
The bait was placed in a simple plastic mesh contain- 
er that allowed the bait scent to disperse as soon as the 
system was placed in the water. The bait station was 
considered to have started when BotCam arrived at the 
seafloor, as determined from the video recording. From 
that point, the cameras were allowed to record for a 
minimum of 30 minutes before BotCam was recovered. 
Data analysis 
Each video stream from the two cameras was viewed 
independently. Each video was viewed in 3-minute inter- 
vals to allow for flexibility in analyzing the data. The 
data from the 10 intervals per 30-minute station could 
be combined into larger intervals or a subset could 
be randomly selected for statistical comparison with 
data from other bait stations. The maximum number 
(MaxNo) of each species seen in any one frame within 
the time interval (Ellis and DeMartini, 1995) and the 
exact time from the start of the deployment to the time 
of first arrival (TFA) of each species seen over the entire 
30 minutes were recorded. Further, the largest MaxNo 
from all the increments was noted as the MaxNo for the 
deployment for each species observed. 
For the purposes of this study, enumeration and mea- 
surements were performed only for the two primary 
bottomfish species of interest, P. filamentosus and E. 
coruscans, which were also the two most frequently 
observed of the “deep 7” species and represent the ma- 
jority of the bottomfish catch in the Hawaiian Islands 
(Haight et al., 1993a; Parke, 2007). 
Bottomfish fork-length measurements were made from 
the video recordings by using a software package called 
Visual Measurement System (SVS) (Geomsoft, Victoria, 
Australia). With this software, the video streams were 
synchronized by time by using the SVS device, and then 
viewed simultaneously frame by frame. Measurements 
of lengths for E. coruscans and P. filamentosus were 
conducted by using the MaxNo video frame and adja- 
cent frames to avoid repeat measurement of individual 
fish congregating around the bait. Each individual fish 
was measured six times from different video frames to 
evaluate the consistency of the measurement technique. 
This method of only measuring at MaxNo may bias the 
data by possibly selecting for smaller schooling fish 
(Willis et al., 2003). 
To specifically test the precision and accuracy of the 
stereo-photogrammetric method of fish measurement, 
a separate experiment was performed in shallow wa- 
ter. BotCam video was used to measure four different 
fish models (foam cutouts shaped like fish) of varying 
length (469.9 mm, 581.0 mm, 628.7 mm, and 997.0 
mm) and body depth. The models were filmed at vari- 
ous locations in the field of view at distances of 3 m 
and 6 m from the cameras. The BotCam was rotated 
by a diver so that the fish traversed the field of view to 
simulate swimming. The models were moved vertically 
to obtain coverage of the models throughout the fields 
of view of the cameras and the models were measured 
at haphazard angles. Length measurements on each 
fish were made by three scientists using stereophoto- 
metric software. 
The relative distributions of each species across sub- 
stratum and slope categories described above were 
evaluated within the framework of a generalized lin- 
ear model based on a Poisson distribution and log-link 
function. The model development for predictor variables 
was based on likelihood ratio tests with a comparison 
of the full and reduced models. A Pearson chi-square 
goodness-of-fit test was used to evaluate the appropri- 
ateness of the model fits (Kutner et al., 2005). Model 
fitting included habitat and depth categories and their 
two-way interaction. 
Results 
Thirty-three sampling trips were conducted between 
June 2006 and February 2007, on which a total of 102 
BotCam deployments were completed. The fabrication 
of a second BotCam system toward the end of the study 
increased the average number of deployments per boat 
trip to 5.5. Six to eight drops could easily be conducted 
per day depending on travel time from port to the deploy- 
ment sites. Of the 102 BotCam deployments, 82 were 
successful and were distributed amongst habitat and 
depth categories as outlined above (Table 1). Of the 20 
that failed, four landed below 300 m so their record- 
ing was too dark; four landed above 100 m outside the 
