220 
Fishery Bulletin 119(4) 
Real-time vehicle altitude data and seafloor depths from 
the shipboard echo sounders are necessary to appropri- 
ately control the vehicle and avoid obstacles. Live views 
from the vehicle are provided by forward- and side- 
facing cameras (AVT Prosilica GT1920, Allied Vision 
Technologies GmbH, Stadtroda, Germany) mounted at 35° 
angles and constantly lit with LED flood lights. Cameras 
collected images at a resolution of 1936 x 1456 pixels and 
at 12 frames per second. The vehicle collected real-time 
hydrological (e.g., salinity, temperature, and depth), vehi- 
cle altitude, and compass data that were stored in onboard 
computers. Transects were surveyed in parallel to the set 
direction of the MOUSS platforms and perpendicular to 
the camera FOV. Between transect runs, the vehicle alti- 
tude was increased to a safe height, and the vehicle then 
was towed in a large oval. Therefore, during surveying of 
transects, the TV always traveled in the same direction 
down the transect line with ~30 min elapsed time between 
each transect. 
The AUV (1.90 x 0.34 x1.50 m, in length, width, and 
height, respectively) operated free from the ship at all 
times during a dive, following a preprogrammed route. 
The AUV was equipped with onboard power, computing, 
acoustic beacons, and an acoustic Doppler profiler for 
bottom-tracking, conducting its operations, and navigat- 
ing through the environment. The vehicle’s stereo cameras 
were oriented to photograph the seafloor (i.e., downward 
facing) and coupled with a xenon camera strobe to capture 
imagery during transit. Environmental data (e.g., salin- 
ity, temperature, depth) and vehicle altitude and position 
data were collected and stored onboard the AUV. At the 
start of the mission, the AUV descended to the seafloor, 
navigated to the programmed transect start position, 
and transited the desired course. Transects were surveyed 
parallel to the set direction of the MOUSS platforms and 
at a preprogrammed constant altitude above the seafloor 
(either 2 or 4 m), in both directions, and at increasing dis- 
tances from the camera with each successive transect. In 
2014, to ensure the vehicle was observed with the MOUSS 
platforms, vehicle altitude was lowered to 2 m; therefore; 
vehicle altitude varied between sites although altitude 
during a dive was held constant. Each successive transect 
was separated by an interval of approximately 30 min 
from the previous transect, similar to the intervals used 
for surveys conducted with the TV. 
Vehicle range from the MOUSS platform was calcu- 
lated by using stereo-camera imagery data when the 
vehicle was observed and then by fitting a linear model 
to estimate values when the vehicle was outside of the 
MOUSS sampling volume. During vehicle passage, 
5 vehicle position coordinates per second were measured 
in centimeters from the stereo-camera origin. We then 
used the point cloud positional data and a linear model to 
estimate vehicle position for periods when the vehicle was 
outside of the stereo-camera FOV. Positional data were 
then used to estimate vehicle range relative to when the 
vehicle passed across the y intercept of the cameras (i.e., 
where x=0) for each second from 1.03 min before and after 
that moment (i.e., up and down range from the camera 
origin). This linear model assumes that the vehicles tran- 
sited in a relatively straight line and that error in posi- 
tion estimates was reduced by obtaining estimates for 
positions of 5 data points per second. Therefore, the total 
time for video annotation was 2.06 min. This time frame 
was selected because the fastest vehicle (the TV) tran- 
sited across the stereo-camera FOV in 8 s and because 
abundance estimates from the MOUSS platforms were 
collected in 4-s bins. 
In this way, several bins of count data can be captured 
during the time when the vehicle was just entering, within, 
or just departing the MOUSS platform sampling volume 
(i.e., the coincident sampling volume). Sometimes MOUSS 
platforms landed on uneven seafloor and were tilted in 
various orientations; as a result, measuring exact alti- 
tude of the vehicle during transit was not always possible. 
Therefore, we collected qualitative data as a measurement 
of vehicle transit altitude in lieu of measuring exact tran- 
sit altitude. Qualitative vehicle altitude above the seafloor 
(RVA) was qualitatively ranked as low (~1—2 m; code: 1), 
middle (~2—5 m; code: 2), and high (>5 m; code: 3). 
Video annotation and relative abundance 
Fish were identified to the lowest taxon possible, and 
attraction and avoidance patterns relative to the vehicle 
were noted. When there were fewer than 50 individuals 
of a species in a given video frame all individuals were 
counted. When there were >50 fish, the total number was 
estimated by subsampling a portion of the school and 
extrapolating the subsample by the total area the school 
occupied on the video screen. 
Videos were annotated in 2 different ways to test the 
following: 1) fish acclimation to stationary cameras after 
first deployment and 2) change in relative fish counts in 
relation to vehicle passage. To evaluate fish acclimation 
during the first hour of deployment, 2-min intervals were 
randomly selected from the first hour of MOUSS platform 
deployment and prior to vehicle deployment. Species- 
specific fish counts during this time were estimated in 4-s 
bins. To analyze fish response to mobile vehicles, species- 
specific fish counts were collected at 4-s intervals during a 
2.06-min video segment for which the midpoint is defined 
as the video frame when the vehicle crossed the stereo- 
camera origin. Depending on vehicle speed, this method 
resulted in fish counts for 1 min prior to and following 
vehicle transit, with a 4-s interval when the vehicle 
passed directly in front of a MOUSS platform. The video 
frames in which the vehicle was captured with the stereo 
cameras were considered the coincident sampling volume, 
and the exposure time in the sampled volume was depen- 
dent on vehicle speed. Because speed varied by vehicle 
and condition, we excluded time as a factor in the models 
in favor of estimating vehicle range to the coincident sam- 
pling volume. 
With a few exceptions, such as greater amberjack (Seri- 
ola dumerili), scamp (Mycteroperca phenax), and gray 
snapper (Lutjanus griseus), specific species were infre- 
quently captured in images coincident to vehicle transit. 
